Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Architectural Transformations in Network Services and Distributed Systems
Andriy LuntovskyyJosef Spillner
Architectural Transformations in NetworkServices and Distributed Systems
Andriy Luntovskyy bull Josef Spillner
ArchitecturalTransformations in NetworkServices and DistributedSystems
Andriy LuntovskyyBerufsakademie SachsenDresden Germany
Josef SpillnerService Prototyping LabZuumlrcher Hochschule fuumlr AngewandteWissenschaftenWinterthur Switzerland
ISBN 978-3-658-14840-9 ISBN 978-3-658-14842-3 (eBook)DOI 101007978-3-658-14842-3
Library of Congress Control Number 2016957988
Springer Viewegcopy Springer Fachmedien Wiesbaden Gmb 2017This work is subject to copyright All rights are reserved by the Publisher whether the whole or part ofthe material is concerned specifically the rights of translation reprinting reuse of illustrations recitationbroadcasting reproduction on microfilms or in any other physical way and transmission or information storageand retrieval electronic adaptation computer software or by similar or dissimilar methodology now known orhereafter developedThe use of general descriptive names registered names trademarks service marks etc in this publication doesnot imply even in the absence of a specific statement that such names are exempt from the relevant protectivelaws and regulations and therefore free for general useThe publisher the authors and the editors are safe to assume that the advice and information in this book arebelieved to be true and accurate at the date of publication Neither the publisher nor the authors or the editorsgive a warranty express or implied with respect to the material contained herein or for any errors or omissionsthat may have been made
Printed on acid-free paper
This Springer Vieweg imprint is published by Springer NatureThe registered company is Springer Fachmedien Wiesbaden GmbHThe registered company address is Abraham-Lincoln-Str 46 65189 Wiesbaden Germany
H
To our caring wives
Preface
About the Book
Book objectives You are reading a book which aims to cover the broad field of recentinnovations in network services and distributed systems The target group of the bookencompasses students of universities and technical high schools graduated engineers aswell as teaching staff If you are somebody else do not worry the covered subjects maystill be of interest to you This book offers its readers a dual functionality
As a monograph With the given work we decided to help not only the readersand students but also ourselves as the professionals who are actively involved inthe networking telecommunications and systems communities by understanding thetrends which have developed in the recent decade in distributed systems and networkingapplications Important architectural transformations of modern distributed systems areexamined and presented in survey style Examples of new architectural solutions fornetwork (mobile) services and applications are discussed Among them are the Internetof Services Clouds Smart Grids Parallel and Distributed Computing Fog Computingand the Internet of Things to mention a couple of popular concepts
As a handbook Current technologies standards and research results for advanced(mobile) networks connected devices and provisioned services as well as for higher-level network functions and software applications are focused within this book from apractical angle The authors highlight how these technical underpinnings to our digitalcommunication and collaboration infrastructure are being transformed to reflect societyrequirements Efficient architectures principles and systems for mobile and wirelesscommunication criteria for optimisation of networks and distributed systems as well ascentral ideas to new system concepts are widely discussed herein Use case presentationsand studies with in-depth technical descriptions along with a test exam strengthen thenature of this book as handbook to use for courses and projects
vii
viii Preface
Learning objectives The learning objectives targeted by the book are as follows
1 Readersstudents should be able to combine integrate analyse and manage thesolutions to the above-mentioned technologies (Clouds Smart Grids Parallel andDistributed Computing Fog Computing Internet of Services Internet of Things) Theyshould also be able to implement custom systems on the basis of an adequate conceptualgrounding in practical projects
2 As a result readersstudents become skilled to create and evaluate well-performingreliable and secure access aspects to data and network applications distributed systemsand mobile apps The systems and services should be usable in a data protection-compliant manner and aligned with user preferences
3 Readersstudents become educated to develop custom architectures of network servicesand distributed systems as well as to comment critically on the associated problems
Numerous examples in the chapters comparison tables excursions into technologicalstacks figures with structures and demonstrations are highlights of this book Everychapter has a list of keywords complemented by actual system examples a summaryand continuing bibliographic records Furthermore at the end there is a whole chapterdedicated to repetition and self-controlling by offering questions and answers to many ofthe discussed topics along with further insight into the research behind the covered systemsand services
Motivation Despite the existence of a broad range of scientific and practical literatureon the topics of distributed computing cloud computing privacy-preserving systemsgreen IT Internet of Things and so forth from our perspective as researchers andlecturers there is a distinct lack of combined monographshandbooks with a pretenceto be useful to education In particular most of the literature describes technologicalsnapshots as points in time Instead we want to explicitly include historical backgroundinformation and focus on the ongoing evolution and trends which are similar in manyareas Furthermore we were not satisfied with literature which merely lists positions andstandards instead of allowing the reader to dive right into the technology by offeringconcrete implementation and use case links Especially for students in co-education (forinstance BerufsakademieDuale Hochschule Fachhochschule and (houmlhere) Fachschule inGermany and Switzerland vocational and community colleges in the US) the practicallinks are essential to decide whether or not a certain technology should indeed be evaluatedfor upcoming projects
The book partially continues the educational approach of a previous book calledPlanning and Optimisation of Computer Networks Methods Models Tools for DesignDiagnosis and Management in the Lifecycle of Wired and Wireless Computer Networksby Luntovskyy Guetter and Melnyk which appeared by Springer Vieweg in Germanlanguage in 2011 The original title is Andriy Luntovskyy Dietbert Guetter IgorMelnyk Planung und Optimierung von Rechnernetzen Methoden Modelle Tools fuumlr
Preface ix
Entwurf Diagnose und Management im Lebenszyklus von drahtgebundenen und draht-losen Rechnernetzen Springer Fachmedien Wiesbaden GmbH 2011 435 pages (ISBN978-3-8348-1458-6) 1st edition 2011 with 245 figures und 64 tables The present bookcomplements and extends the range of topics It addresses the evolved development fromcomputer networks to network-integrated and network-connected services in particularcloud and fog services as well as modern architectures of distributed (mobile) applicationssuch as 5G and low-energy radio links The new book therefore presents a holistic view ontransformation processes which are nowadays often less technically motivated but ratherby the needs of the society which is subject to a higher degree of pervasive services Thebenefits for society are about ecology (green networks) privacy (secure clouds) comfort(always on) and economy (pay as you go)
Structure of the Book
This book is divided into seven chapters The first chapter offers a birdrsquos perspective onthe history and present development of networking and service topics The second chapterpresents state-of-the-art distributed systems and uses them to explain the architecturaltransformations which most of todayrsquos systems are subject to In the chapters three to sixdifferent architectures and systems will be presented including clusters clouds fogs andmobile applications The seventh chapter offers a holistic view on security in networkedservices Finally five appendices and one more auxiliar digital appendix complete thebook
bull Chapter 1 ndash Periodisation of Network Service Development The evolution of hardwareand infrastructure on one hand and of services on the other hand is divided into fourphases each
bull Chapter 2 ndash Architectural Transformation in Distributed Systems Clusters and cloudspeer-to-peer architectures and distributed databases will be presented and reflected onin the context of the evolution and transformation of systems
bull Chapter 3 ndash Evolution of Clustering and Parallel Computing Clusters grids andparallel computing will be introduced Their benefits concerning the performance ofcomputing but also the necessary trade-offs with energy consumption and price willbe highlighted The management of resources and applications in these environmentswill also be explained
bull Chapter 4 ndash Cloud Computing Virtualisation RAICs and SDN This chapter willintroduce contemporary cloud stacks and services including programmable networksvirtual teleconferences and safe data backups
bull Chapter 5 ndash Smart Grid Internet of Things and Fog Computing Beyond the softwareside small connected hardware devices and the connection between computer networksand energy distribution networks will be covered in this chapter
x Preface
bull Chapter 6 ndash Future Mobile Communications From 4G to 5G 5G Enabling TechniquesMobile communication protocols for global (phones) and local distances will bepresented A special focus is on the upcoming 5G connectivity
bull Chapter 7 ndash Security in Distributed Systems This chapter will give a holistic view onwhat is commonly called security by introducing into concrete protection goals andmatching security layers It will also include a discussion of privacy and legal aspectswith a focus on how users can protect their activities and communication in todayrsquos andtomorrowrsquos distributed systems
bull Appendices First selected originators and designers of distributed systems will bebriefly presented Then specific research projects with recent results which contributeto the evolution and transformation will be introduced The further parts containexplanations to common acronyms in mobile and wireless technologies a repetitionand control part to track the learning progress when reading the book and finally anexample of a written exam to the discussed subjects The solutions to the exam areavailable as auxiliar digital appendix
Dresden Germany Andriy LuntovskyyWinterthur Switzerland Josef Spillner
Acknowledgement
All our graceful heartrsquos acknowledgements to Prof Dr rer nat habil Dr h c AlexanderSchill (encouragements and challenges) Dr rer nat Dietbert Guumltter (proofreading) ProfDr Andreas Westfeld Prof Dr Thomas Horn Dr Reiner Keil (inspiration in absentia)and many other colleagues students and reviewers for their helpful and friendly supportthe inspirations and co-operation while completing this work
Our special acknowledgment goes to Dr-Ing habil Igor Melnyk for his altruisticcontribution to the modelling of the waste heat and cooling process in ldquogreenrdquo data centersand clouds
xi
About the Authors
The book contents have been primarily provided by Andriy Luntovskyy Some sectionsand editorial guidance were provided by Josef Spillner Most of the material is publishedfor the first time although some is based on previous research papers including jointpapers by the authors and material kindly added by fellow academics
Andriy Luntovskyy Prof Dr habil
Andriy Luntovskyy is with BA Dresden University of Cooperative Education DresdenGermanyOffice Room 2105 Hans-Grundig-Strasse 25 01307 Dresden (Johannstadt) GermanyPhone +49 (0)351-44722-703Fax +49 (0)351-44722-9520Email AndriyLuntovskyyba-dresdendeWWW httpwwwba-dresdendeWWW (EN) httpsitesgooglecomsiteluntovskyyWWW (UA) httpsitesgooglecomsiteandriyluntovskyyWWW (DE) httpwwwba-dresdendedestudiumstudienangebotitansprechpartnerhtml
xiii
xiv About the Authors
Andriy Luntovskyy is member of the Academy of Sciences for High School of Ukraine(ANVSUorgua) and member of the Academy of Telecommunications of Ukraine andInternational IT Academy
Teaching and Classes Computer Networks Mobile Communication and TelematicsBasics of Programming and Software Technology Distributed Systems Operating Sys-tems Web-Applications and Office Communication Data Security and IT Legacy Basicsof Computer Science and Business Informatics Guest lectures in Ukraine and Polandclasses for bachelor master and PhD students
Research CANDY ndash Computer-Aided Network Design utility Design of WiredWireless and Mobile Networks Clouds Clustering and Mobile Computing Web ServicesSOA and Virtualisation Methods Mobile and Wireless Networks Energy Efficiencyin Networks Wireless Sensor Networks Smart Grid and IoT Multiservice MobilePlatforms
Attendance and co-chairman at multiple conferences and forums (CEBIT 2007 20082011) Publications two books are published in Germany (2008 2011) other 12 booksin mother tongue in Ukraine more than 130 papers to conferences and magazines amongthem multiple IEEE Xplore publications
Josef Spillner Dozent Dr-Ing habil
Josef Spillner is with Zurich University of Applied Sciences (ZHAW) School of Engi-neering Winterthur SwitzerlandOffice Room O317 Obere Kirchgasse 2 8400 Winterthur SwitzerlandPhone +41 (0) 58 934 45 82Fax +41 (0) 58 935 45 82Email josefspillnerzhawchWWW httpwwwzhawch=spioWWW httpwwwserviceplatformorg
Josef Spillner performs research on service and cloud ecosystems is the initiator ofthe Open Source Service Platform Research Initiative founder of the Cloud Storage
About the Authors xv
Lab at Technische Universitaumlt Dresden in Germany (TUD) and the head of the ServicePrototyping Lab at ZHAW
Teaching and classes Introduction into Research Areas of Computer Science Devel-opment of Distributed Sysstems on the Basis of SOA Complex Internship for Service andCloud Computing OS and Computer Networks Basics of Programming and SoftwareTechnology Distributed Systems Python Programming Classes for bachelor and masterstudents as well as non-IT students in particular media informatics and industrialengineers
Research THESEUSTEXO ndash New Technologies for the Internet of Services fundedby the German Ministry of Economics (BMWi) FlexCloud ndash Flexible Service Archi-tectures for Cloud Computing funded by the European Social Fund (ESF) DaaMobndash Service-oriented Platform Concepts for Cross-System Third-Party Applications withMobile Components in the Internet of Things funded by the German Research Council(DFG) Further research on XML Schema Web Service GUIs Cloud Controllers CloudCockpits and Energy Efficiency Stealth Computing
Attendance and involvement with multiple conferences and workshops Publicationsbooks co-authorship more than 50 papers and journal articles technical reports with HPIFuture SOC Lab IEEE and ACM conference chairing
List of Abbreviations
2PC Two-Phase Commit Protocol 26ndash28 35ndash37 40 422PL Two-Phase Lock 37 42
ACID Atomicity Consistency Isolation Durability 26 28 30 35 38 40AEF Advanced Evasion Firewall 247 272AES Advanced Encryption Standard 255 257 258 265 276 277 283API Application Programming Interface 46 69 81
B2B Business-to-Business 23BOINC Berkeley Open Infrastructure for Network Computing 45 47 61 62 66ndash70
C-S Client-Server 13 19 20 22ndash24 43CAD Computer-Aided Design 23 135 177 178CDB Central Database 13 30ndash33CIDN Collaborative Intrusion Detection Network 247 268 271 273ndash276
DB Database 29 30 32 34 38DDB Distributed Database 13 19 30ndash38 42 43DDoS Distributed Denial of Service 2DIDO Distributed Input Distributed Output 211 225 230 241ndash244DNS Domain Name System 15DSL Digital Subscriber Line 2 3
EAI Enterprise Application Integration 81 82 84 98EM Electro-Magnetic 138 139 141 168 174 208ERE Energy Reuse Efficiency 136ESB Enterprise Service Bus 16
xvii
xviii List of Abbreviations
FLOPS Floating-Point Operations Per Second 46ndash50 54 59ndash61 67FUSE File System in Userspace 124
GSM Global System for Mobile Communications 211 212
HPC High-Performance Computing 45 59 66HSDPA High Speed Download Packet Access 211 213 217HTTP Hyper-Text Transport Protocol 80 82ndash84 90HVAC Heating Ventilating and Air Conditioning 8 9
IaaS Infrastructure-as-a-Service 9 77 79 81 85ICMP Internet Control Message Protocol 8 140IDS Intrusion Detection System 247 270 271 273 274 276IETF Internet Engineering Task Force 6IMS IP Multimedia Subsystem 213ndash216 225 241IoS Internet of Services 1 3 4 18 77 79 81 85 113 135 183 184 187 188IoT Internet of Things 1 4 5 9 10 135 159 168 184 185 187ndash194 196 203
207 208IP Internet Protocol 5 8 140 160 178 180 192 213 214 217 225 227 235
244 259 260 264ndash272 276 277 281 283 297IPS Intrusion Prevention System 247 270ndash273 276ISDN Integrated Services Digital Network 1 6
KNX KNX Home and Building Control Standard 7 9 140
LAN Local Area Network 8 19 140 159 174 176 178 195 198 201LEACH Low-Energy Adaptive Clustering Hierarchy 166LON Local Operating Network 7 9 140LTE Long-Term Evolution 211 213 214 223 225 226 237 244
MAC Media Access Control 144 161 163 164 168 170 172ndash174 187 189MCM Majority-Consensus-Method 37MIMO Multiple Input ndash Multiple Output 213 225 234 237 240 241MIPS Million Instructions Per Second 48 49
NAS Network-Attached Storage 113NIST National Institute of Standards and Technology USA 18 79 80 85 114 145NTP Network Time Protocol 15
OFDM Orthogonal Frequency-Division Multiplexing 159 213 239 240OFDMA Orthogonal Frequency Division Multiple Access 239
List of Abbreviations xix
OS Operating System 25 26 190 195 196 198 200 208OSGi Open Services Gateway Initiative 15OSI Open Systems Interconnect 144 145 259 261 272
P2P Peer-to-Peer 13 19ndash23 43PaaS Platform-as-a-Service 9 77 81 85 86 89PCS Primary-Copy-Schema 37PEV Plug-in (Hybrid) Electric Vehicles 138 140 141 149PGP Pretty Good Privacy 247 260 262 276 277 289PLC Power Line Communication 135 148 158 159 189PoE Power over Ethernet 9PUE Power Usage Effectiveness 3 9 136 150 151 153ndash155 157 158
QoE Quality of Experience 85 86QoS Quality of Service 1ndash4 17 77 79 82 85 86 110 113 114 138 154 161
166 197 208
RAIC Redundant Array of Independent Clouds 77 91 111 113 119ndash123 125ndash131RAID Redundant Array of Independent Disks 112 113 119ndash122REST Representational State Transfer 82ndash85 89RFC Requests for Comments 6 213 214RSA Rivest Shamir Adleman Cryptosystem 255 257 258 265 276 282 283
SaaS Software-as-a-Service 9 77 79 81 85 89 92SAN Storage-Area Network 81 112 113SDN Software-Defined Networking 77 92 105ndash110 225 230 232SET Secure Electronic Transaction 279 281 283 284 287 288SIF Stateful Inspection Firewall 247 270ndash272SIP Session Initiation Protocol 213 214SLA Service Level Agreement 2 81 82 85 91SME Small and Medium Enterprise 7 139SMLIF Stateful Multi-Layer Inspection Firewall 247 272 276SMP Symmetric Multi-Processing 60 61SMTP Simple Mail Transmission Protocol 6SNMP Simple Network Management Protocol 8 140 192SOA Service-Oriented Architecture 79 82ndash84 88 89 98 113SOAP Simple Object Access Protocol 83 85 90SQL Structured Query Language 30 35 36 39 40SSL Secure Sockets Layer 264ndash266
xx List of Abbreviations
TLS Transport-Layer Security 247 256 258 260 264 265 267 268 279 281ndash283 287 288 303
UMTS Univeral Mobile Telecommunications System 211 213 214UPnP Universal Plug and Play 15
VM Virtual Machine 85 92ndash95 98 100 105ndash108 114VoIP Voice over IP 214 215VPN Virtual Private Network 247 265ndash268 270 271 281 283 302VTEO Virtual Telecommunication Engineering Offices 77 84 85 88ndash91
W3C World Wide Web Consortium 6WAF Web Application Firewall 247 270 276WAN Wireless Area Network 145 159WiMAX Worldwide Interoperability for Microwave Access 139 149 159ndash161 178
188WLAN Wireless Local Area Network 8 19 140 159 161 171 178 180 187 195
202 211 212 214 224 225 234 240 241 243 244WPAN Wireless Personal Area Network 19 135 158 168WSN Wireless Sensor Networks 139 141 161ndash166 173 174 189
XaaS Everything-as-a-Service 79XMPP Extensible Messaging and Presence Protocol 70 73 74 85
Contents
1 Periodisation of Network Service Development 1References 10
2 Architectural Transformations in Distributed Systems 1321 Software Architectures and Communication Patterns 1322 Distributed Service Systems Clustering Grids and Clouds 1723 Architectures Peer-to-Peer 1924 Performance Optimisation 2325 Distributed Transactions 2626 Distributed Databases 3027 System Examples Google Spanner a Global DDB 3828 Conclusions 43References 44
3 Evolution of Clustering and Parallel Computing 4531 Clustering and Grids Performance Parameters and Basic Models 4832 Performance-Energy-Price Trade-Offs in Clusters and Grids 6233 Resource Management in Clusters 6434 Application Management in Clusters 6535 Application Management in Grids 6636 Distributed Applications 7137 Conclusions 74References 75
4 Cloud Computing Virtualisation Storage and Networking 7741 Clouds Technology Stack Basic Models and Services 7842 Virtualisation of Services and Resources 9243 SDN ndash Software-Defined Networking 10544 Backup Services within Clouds as Advanced Cloud Backup
Technology 110441 Backup as Important Component of Informational Safety 111
xxi
xxii Contents
442 RAIC Storage Service Integration 11745 RAIC Integration for Network Storages on Mobile Devices 125
451 Efficient Access to Storage Services from Mobile Devices 126452 A New Must-Have App RAIC Integrator for Smartphones 128
46 Conclusions 131References 131
5 Smart Grid Internet of Things and Fog Computing 13551 Smart Grid as Integration Technology for the Networks of
Energy Supply and Telecommunication 136511 Services Architectures and Multi-level Models 144512 Smart Grid Enabling Network Technologies 158513 Case Study A CAD Toolset for the Design of
Energy-Efficient Combined Networks 17752 From Internet of Services to Internet of Things Fog Computing 184
521 Enabling Technologies for IoT 188522 Case Studies on IoT with On-Board Micro-controller
Raspberry Pi 194523 The Future Industry 40 Vision 203524 Fog Computing 204
53 Conclusions 206References 209
6 Future Mobile Communication From 4G To 5G 5G EnablingTechniques 21161 Conventional Techniques 211
611 LTE Networks 213612 Satellite-Based Radio Systems 215
62 A New Generation of Mobile Communication 222621 Visions and Requirements 224622 5G Inter-Operability 233623 Future Standard IMT 2020 Deployment Scenarios 235624 Resource Allocation Method for Future WLAN 241
63 Conclusions 244References 244
7 Security in Distributed Systems 24771 Security and Protection Goals 24872 Protection Techniques 253
721 Checksum and Digest 254722 Encryption 255723 Steganography 258
Contents xxiii
724 Orchestration Parallelisation and Multiplexing 258725 Anonymisation 258726 Trusted Computing and Physical Protection 259
73 Security Layers 259731 Network Encryption IPsec 259732 Transport Encryption TLS 260733 Content Encryption SMIME and PGP 260734 Authorisation Kerberos and OAuth2 261735 Further Secure Services DNS-SEC VPNs and Proxies 261
74 Security Protocols and Network Concepts 26175 Firewalls 26876 Security in Web Applications Legal and Technological Aspects 279
761 Technological Aspects of Data Security GuaranteeingWeb Systems 281
762 Legal Aspects of Data Security Guaranteeing Web Systems 28377 Steganography in Distributed Systems 288
771 Steganography in Development 290772 Steganography Main Concepts 294773 Watermarks and Steganography 298
78 Anonymity and MIX Networks 30179 Conclusions 306References 307
Appendix A Selected Originators and Designers of Distributed Systems 309A1 Edgar Frank ldquoTedrdquo Codd 309A2 Tom De Marco 310A3 Grady Booch 310A4 James Gosling 311A5 Sir Timothy John Berners-Lee 311A6 Tim OlsquoReilly 312A7 Roy Thomas Fielding 313A8 Sergey Brin 313A9 Philip R Zimmermann 314A10 Remembering the Pioneers 314
Appendix B Research Focus 317B1 CANDY Network Planning 317B2 FlexCloud Flexible Architectures for Cloud Computing 319B3 DaaMob Service Platform Data Service Management 319
Appendix C Acronyms for Mobile and Wireless 323
Appendix D Repetition and Control of Learning Progress 327D1 New Generation (Mobile) Networks 327
xxiv Contents
D2 Periodisation of Computer Networks Phases I to IV SmartGrid IoT and Fog Computing 328
D3 Architectural Transformation in Distributed Systems 328D4 Cloud Computing 329D5 Virtualisation Concepts 330D6 Performance Characteristics of Digital Computers
Performance Optimisation in Distributed Systems 331D7 Distributed Computing Parallel Computing and Acceleration Models 331D8 Towards 5G 332D9 Security Aspects in NGN 332D10 PGP and Steganography 334
Appendix E Example of a Written Exam to the Discussed Subjects 337
Index 343
1Periodisation of Network Service Development
Keywords
Networks bull Services bull Quality of Service (QoS) bull Internet of Services (IoS) bullClouds bull Smart grid bull Internet of Things (IoT) bull Fog computing
Information and communication technology is moving fast What are grids for nowadaysIs anybody still using Integrated Services Digital Network (ISDN) connections Willthe lsquodigital fogrsquo be around all of our devices and for how long when on batteries Whatis the cost of safely storing one digital photo taken on the mobile phone for the rest ofour lifetime Readers who have immediate answers to such questions are asked to putthis book aside and spend their time with more pleasure All other readers are howeverinvited to follow us briefly through the history of network services and distributed systemsthrough the past transformations and current trends in order to learn about the rathercomplex landscape of distributed service systems in the future These digital physicaland combined (cyber-physical) systems affect our daily lives as we interact with themthrough screens and devices software applications processes and ambient sensors
Technology development in four phases Network services and distributed systems aretwo pillars of the same trend To make application functionality provided from singlecomputers or millions of connected devices available to billions of people Internet andweb applications including online social networks and digital telephony already todayneed to scale to billions of users which would be impossible on a single machineInstead many computers are clustered and many clusters are geographically dispersedand connected so that users perceive them as single service The perception is trained forhigh performance high reliability high privacy and security low cost low effort and lowenergy consumption among other factors Services not offering all of these benefits will
copy Springer Fachmedien Wiesbaden GmbH 2017A Luntovskyy J Spillner Architectural Transformations in Network Services andDistributed Systems DOI 101007978-3-658-14842-3_1
1
2 1 Periodisation of Network Service Development
have decreasing chances to compete for users and will ultimately fail to be sustainableTrust and reputation would in such cases be hard to recover
It took computer scientists and the IT industry many years to achieve the breakthroughtowards this vision In the course of development of networked applications and servicesincluding telecommunication web and cloud services offered on-demand in any situationfour distinct phases in the technological foundation can be identified
The first phase starting with the roll-out of networks and the Internet (about 1970ndash2000)had the purpose of offering the functionality and of ensuring improvements to the QoSThe QoS considerations were mostly confined to strict technical network characteristicswithout taking end-to-end user experience into account Bandwidth increased and latencydecreased To put the bandwidth development into perspective In 1999 a 56 kbits modemconnected to copper telephony networks was the norm for private users and just about tobe replaced by faster Digital Subscriber Line (DSL) connections with about 768 kbitsdownstream bandwidth Consumers could only rely on such numbers as upper bounds ina best-effort service market and could not easily translate these numbers into applicationbenefits for instance video quality or file transfer performance
In the enterprise market large computing centers were economically effective dueto using broadband Internet connections which enabled the consolidation of a lot ofcompute and storage resources behind a single data pipe They helped also in mitigation ofDistributed Denial of Service (DDoS) attacks due to load distribution between severalservers and links The system reliability was improved due to better availability of spareparts (hard drives power units switches etc) the employment of redundant units whereverpossible and emergency power generators in large centers where they were feasibleSimilarly the application availability and scalability was increased with replicated setupsin high-availabilityfailover and load-balancer setups respectively
Ultimately the phase has been about connecting people to the Internet in other wordsan Internet of People A simple formula characterises the first phase
GoalPhase1 WD MaxQoS (11)
In the second phase of development of Internet services (about 2000ndash2010) theimprovement of QoS was accompanied by explicit cost optimisation among otherreasons due to hardware consolidation and server virtualisation in combination with QoSguarantees codified in a Service Level Agreement (SLA) These mandated a minimumcost by strictly given QoS constraints But also the large size of computing centers still ledindirectly to less cost on the side of customers due to the economy of scale when buyinglarge charges of spare parts and electricity The maintenance cost in the large computingcenters is also less than in smaller ones because the servers are updated centrally withsecurity patches upgrades can be better tested before deploying and the maintenanceactions are mostly the same at homogeneous servers To give an example The e-commerceseller Amazon had a revenue of about seven billion US$ in 2004 The capacity needed tooperate this business at that time is nowadays added daily to their computing infrastructure
1 Periodisation of Network Service Development 3
It is not yet clear how to compare the technical characteristics of data centres but justlooking at their dimensions demonstrates the trend towards consolidation The LakesideTechnology Center in Chicago one of the largest multi-tenant centres has a usable surfaceof more than 100000 m2 across several floors of a historic printing house MicrosoftrsquosDublin data centre is roughly half this size [10] Major service operators have expandedvastly during the second phase and now operate multiple of such large data centres
On the network side in 2009 16 Mbpss ADSL connection were widely availablein many urban areas in developed countries and even 55 Mbpss VDSL2 connectionswere available in selected areas whereas in 2014 vectoring-based VDSL brought upto 100 Mbpss downstream and 40 Mbpss upstream bandwidth to consumers A slow-down in connection speed growth becomes evident Furthermore the promise of manygovernments during this time to achieve 100 broadband coverage had (and still has)not been achieved anywhere Enhancing the role of hosted applications (in so-calledclouds) as integration path and cost reduction driver for applications and computing powercharacterises this second development phase Consequently an Internet of Services (IoS)in particular cloud services characterises the second phase
GoalPhase2 WD MaxQoS^
Cost Constraints (12)
The third phase (after 2010) was triggered by the trend of ldquogreenrdquo IT and increasingenergy demand and prices The computing centers were built more often in colder regionsof the earth More energy-efficient hardware was installed and software was written withenergy efficiency in mind Processors gained dynamic voltage and frequency settingsamong other techniques which helps shrinking the power consumption over all idleperiods The metric Power Usage Effectiveness (PUE) has gained prominence andconsumers are increasingly aware and demanding of sustainable IT The use of mobilephones to host applications and even mobile services strengthens the awareness due tolimited handset battery capacity Smart grids installations are on the rise and lead to greaterenergy autonomy by turning consumers into providers Therefore to characterise the thirdphase in a formula
GoalPhase3 WD MaxPUE^
QoS QoSmin
^Costs Costsmax (13)
As a by-product of the awareness similar to transportation companies which can alsobe viewed as a public utility the first data centre and hosting businesses have announcedto have met a 100 renewable energy goal [3] This has led to a voluntary green energymarket which in the USA alone has around five million customers who have purchaseddirectly or indirectly approximately 74 million MWh of power generated from renewablesources [6] In Switzerland around 10 of all power consumption is linked to the variousforms of IT an equivalent of 400000 cars in terms of fossil fuel and an increasing numberof providers advertise their decision to contract 100 renewables [2]
4 1 Periodisation of Network Service Development
Fig 11 Periodisation of network service development
Finally the fourth and last phase which has already started but will cause a high impacton computing in the near future needs to be discussed Therefore this book is dedicated tothis phase without dismissing the earlier ones Figure 11 puts all three already identifiedphases with the not yet covered last one into context
The fourth phase the next development vector is about to happen now This phaseis oriented not just at networking services and distributed software applications but to atruly user-focused IoS in many domains It happens across clouds in the frame of the IoTwith many connected small (sometimes wearable) devices cyber-physical systems androbots next-generation mobile networks and ultimately fog and wearable computing Thiscombination expands the always-on always-available pay-as-you-go utility and cloudcomputing paradigm with intelligent network nodes (eg radio network edges smartrouters or even smart watches) and enables via this extension a set of new applicationsand services The features of such an interpretation of fourth-phase computing are asfollows
bull low-latency location-aware energy-efficient use of heterogeneous hardware fromlarge-scale computing centres to tiny nodes
bull very big number of hardware nodes and their mobility based on IPv6 connectivitybull wide geographical distribution of miniaturised hardware self-updating software and
large volumes of databull leading role of wireless access to connect nodes and users even over longer distancesbull service interfaces streaming and real-time applications with guaranteed QoS proper-
ties
1 Periodisation of Network Service Development 5
Fig 12 Fog computing vision (background photo Claudia Jacquemin JOTT Fotografie Dresdenthe depicted place CADCAM system at BA Dresden ndash University of Cooperative Education)
A wider interpretation of fog computing offers the appropriate platforms for IoT cloudsand the smart grid (Fig 12)
According to Eric Schmidt at that time CEO at Google at the World EconomicForum in Davos Switzerland in 2015 ldquoI will answer very simply that the Internet willdisappear There will be so many Internet Protocol (IP) addresses so many devicessensors things that you are wearing things that you are interacting with that you wonrsquoteven sense it It will be part of your presence all the time Imagine you walk into a roomand the room is dynamic And with your permission and all of that you are interacting withthe things going on in the room A highly personalised highly interactive and very veryinteresting world emergesrdquo [7]
This industrial development is bound to happen as so far the miniaturisation ofhardware is still advancing rapidly On the other hand researchers also look into waysto keep the user in the loop and ultimately also in control something typically neglectedby industrial development Therefore new methods for informational self-determinationand manageability of personal devices and services need to be found A typical exampleis a safe networking kill-switch to prevent any communication from a device something
6 1 Periodisation of Network Service Development
found only occasionally on devices despite its usefulness along with a definite off-switchBefore going into the details about the future development the same four phases shall beanalysed from a service perspective
Network services in four phases Along with the technical improvements in serversdevices and connectivity the offered services themselves have evolved over time Onedifference when compared to the hardware technology is the fact that new services almostalways complement existing ones instead of replacing them While it would be hardto order an ISDN connection or a Fiber Distributed Data Interface (FDDI) connectionnowadays we still communicate via decades-old e-mail protocols and locate services viaanother decades-old domain naming protocol
In the first phase (1970ndash2000) basic network services and early web applications werecreated Many network services were and indeed still are defined by an internationalcommunity called the Internet Engineering Task Force (IETF) in public and well-editedRequests for Comments (RFC) [9] An example would be an e-mail sending service(Simple Mail Transmission Protocol (SMTP)) first defined in RFC 821 by Jonathan BPostel in 1982 and subsequently updated to RFC 2821 in 2001 and finally RFC 5321 in2008 Other examples include real-time messaging file transfer and authentication Earlyweb applications include e-commerce shops along with search engines and online news-papers for instance bookscom in 1992 yahoocom and spiegelde in 1994 amazoncomand nytimescom in 1995 and googlecom in 19971998 Their growth in popularity wasmainly driven by the first web browsers as client applications including Mosaic (1992)Netscape Navigator Microsoft Internet Explorer and Opera (all around 1994)
The first phase also contained the first monopolisation tendencies Whereas previouslynetwork protocols were defined and then implemented by multiple vendors especiallyweb applications emerged whose interaction was neither well-known nor easily reim-plementable Web pages as interaction part of web applications were standardised byanother entity the World Wide Web Consortium (W3C) but filled with vendor-specificextensions which even today still cause trouble and processing overhead
In the second phase (2000ndash2010) due to faster home connection speeds peer-to-peerfilesharing applications became popular between consumers An early example has beennapstercom which ceased to exist in the year 2000 only to be replaced by open proto-cols including Bittorrent from 2001 on Other peer-to-peer applications quickly gainedpopularity including video conferences and in the year 2009 the cryptocurrency BitcoinInterestingly some applications such as permanent file storage have mostly remained withcentralised data centres despite peer-to-peer applications being available [1]
Web applications were further growing by faster and more powerful web browserswhich emerged after a perceived innovation poise The browsers were Apple Safari (2002)Mozilla Firefox (2004) and Google Chrome (2008) which turned increasingly into aplatform with all of the associated lock-in and vulnerability issues
In the third phase (since 2010) commercial global-scale services have been competingfor marketshare Online social networking services like facebookcom and twittercom
1 Periodisation of Network Service Development 7
Fig 13 Scheme of services and supporting hardware technology for a single distributed application
claim hundreds of millions of active users which are handled by a global network ofdistributed data centres Millions of devices and sensors are connected to enable moreservices And computing infrastructure services with compute storage and networkingservices have emerged in multiple forms and concentrate applications and services inshared data centres During this time consumers have become increasingly aware of whereservices are hosted and how they are delivered In particular privacy issues have emergedand are not solved yet [5] Figure 13 contains a scheme of todayrsquos distributed networksand services and how consumers interact through and with them
Now we can only speculate which novel services will be enabled by the current waveof technological development This will depend in large part on the knowledge skills andfacilities to enact new services by individual developers and businesses The followingthree fictive scenarios illustrate the hypothesis about the advancement of technologicaltrends in the fourth phase of the chosen periodisation They will be picked up in the nextchapters and illustrated with concrete examples
Scenario 1 Smart grid in an SME What will be a middle-class network connectionfor an Small and Medium Enterprise (SME) in 2020 Only one cable or wirelesslink will provide the utility services such as electricity telephony Internet digital high-definition television and cloud services Room heating will be realised via derivation andrecycling of redundant energy from multiple (virtual) servers The wired and wirelessautomation of local-area as well as piconets like Local Operating Network (LON)KNX Home and Building Control Standard (KNX) ZigBee EnOcean will be used to
8 1 Periodisation of Network Service Development
serve and control the in-door climate Management of such integrated networks can beperformed through Ethernet Local Area Network (LAN)Wireless Local Area Network(WLAN) links as well as convenient protocols like IP Internet Control Message Protocol(ICMP) Simple Network Management Protocol (SNMP) The program supportconfiguration and tuning of the intelligent network is realised with the use of mobiledevices (smartphones and tablets) mobile applications and through offered web servicesrunning in a cloud environment This leads to a smart environment in which all companydevice capabilities are used in combination to their full extent to ensure autarky with highsecurity and privacy but still on-demand scalability beyond the companyrsquos realm and highenergy efficiency with inclusion of all local energy sources and joint brokering of powerand computing supplies We name the outcome of this scenario a smart grid environment
Scenario 2 Energy recycling in data centers Due to use of todayrsquos powerful high-end servers within the contemporary data centers with the installed broadband opticallinks (eg Fibre Channel) a significant amount of heat stands out as a harmful by-product Some companies occupy themselves already with the mentioned problem andare developing their own solutions for the disposal of heat excesses for domestic heatingand air-conditioning facilities the so-called HVAC Among them are hybrid cloud andheat product providers [8] These companies have a portfolio of several correspondingproducts and solutions (Fig 14) inter alia there are cloud infrastructure and platform
Fig 14 Hybrid cloudheat providers combination of smart grid clouds and HVAC
1 Periodisation of Network Service Development 9
services and heat products representing an own smart grid with inter-connected servicesThe clients use the in-door located services of virtual computing centers standardisedcloud services like Infrastructure-as-a-Service (IaaS) Software-as-a-Service (SaaS)and Platform-as-a-Service (PaaS) Among them there are popular applications causinga significant amount of heat from computing services powered by cloud stacks virtualisedoperating systems and add-on services like databases and cron jobs Redundant heat as aldquoby-product of processingrdquo is withdrawn via servers in 1900-racks in the energy storagewhich provides circulation of hot water in the pipes within a building and heating ofpotable water The central system for HVAC facilities is supported via use of Power overEthernet (PoE) as well as wired and wireless automation local-area and piconets likeLON KNX ZigBee EnOcean The mentioned technical solution provides a lower PUEvalue down to 105 or correspondingly an efficiency 1PUE up to 95 compared withthe conventional gridcloud-solutions where it is necessary to remove the excess heat asby-product to install more air-conditioning devices and provide them with power supply
Similarly a growing number of data centres world-wide are inter-connected withmunicipal utility providers to funnel their excess heat into pipes which lead to centralheating systems of housing areas Interesting installations exist in Helsinki Finland whereservers located beneath the Uspenski cathedral in the AcademicaTelecity Group servercentre heat 500 homes as by-product More servers located in a shielded building insideanother building a former electricity station now hosting the Suvilahti data centre evenoffer heat and warm water for 4500 households
Scenario 3 Low-cost and energy-efficient on-board microcontrollers for pico-services But none of the above-mentioned computing systems is energy-efficient enoughto meet the ambitious goals set by environmentalists and to some degree even politicalagendas Switzerland for instance is committed to reduce the emissions in 2030 to just50 of those in 1990 Germany intends to reduce emissions until 2020 to 60 Theelectricity consumption in data centres is in the MWh area and even for tiny computationsa power-hungry large machinery of hardware and support processes is needed Energy-efficient solutions can be provided via small low-cost and low-energy on-board processorson which pico-services such as lambda services are executed on demand The electricityconsumption gets reduced to the kWh area or even less Low-energy home intelligentnodes (3ndash10 W) for private cloud solutions file servers web servers multimedia homecentres etc can be placed on the low-cost energy-efficient on-board microcontrollerslike Arduino Raspberry Pi or Intel Edison as a trade-off solution They offer a cheapalternative and symbolise a step-by-step shift to the IoT But in order to maximise theirpotential an appropriate service and application platform will be needed
An appropriate solution will be the Raspberry Pi on-board-microcontroller (firstdeployed in 2011 in Cambridge UK) with only credit card dimensions in a pod likea matchbox and with the following characteristics [4] A 700 MHz processor a modestamount of main memory up to 1 GB external storage on an SD card an Ethernet connec-tion or a wireless link through a USB dongle and around 35ndash5 W power consumption
10 1 Periodisation of Network Service Development
Naturally there are a lot of scenarios on economical network nodes For instance fora so-called Multimedia Home Centre with the following characteristics a cheap and low-energy Raspberry Pi can be typically used
bull SD-Card as a hard drive with 32 GByte capacity and Raspbian loaded as operatingsystem
bull Multimedia environment XBMC Media Centerbull Multiple audio and video formats (codecs) as well as low power
The newest Raspberry Pi 2 Model B acts as a mini-PC with 6 times the CPUperformance due to a tact frequency of 900 MHz and a quad-core architecture beingoriented to the Windows Developer Program for IoT But even more energy-efficientboards are upcoming including the Genuino with the Intel Curie chip and the Pine A64which even runs on a 37 V Lithium battery
How to read on This was a quick chapter The next ones will have more depth asthey convey the actual knowledge about the mentioned areas In the second chapter thedevelopment of network systems will be summarised and presented with historical andcontemporary systems In the third chapter clusters and parallel computing will be focusedon Virtualised systems and clouds will follow in the fourth chapter Chapter number fivewill step into the physical world and contains information about smart grids smart thingsand smart fog While the sixth chapter will present mobile communication trends the finalseventh chapter talks about security aspects in a broad meaning With such a spectrum oftopics the reader should then be able to understand both old and new large-scale systems
References
1 Bence Bakondi Peacuteter Burcsi Peacuteter Gyoumlrgyi Daacutevid Herskovics Peacuteter Ligeti Laacuteszloacute MeacuteraiDaacuteniel A Nagy and Viktoacuteria Villaacutenyi A P2P Based Storage System with Reputation Pointsand Simulation Results In Central European Conference on Cryptology (CECC) BudapestHungary May 2014
2 Markus Bloesch netrics uumlbernimmt Umweltverantwortung Cloud Computing und Hosting ausDatacenter mit Oumlkostrom aus dem Wasserkraftwerk Hagneck online httpswwwnetricsch20151203cloud-computing-hosting-mit-oekostrom 2015
3 Alisa Davis Equinix Goes 100 Renewable with 225-MW Wind Energy Purchase onlinehttpapps3eereenergygovgreenpowernewsnews_templateshtmlid=2082 2015
4 Raspberry Pi Foundation Raspberry Pi Hardware online httpswwwraspberrypiorgdocumentationhardwareraspberrypiREADMEmd 2015
5 Thomas Loruenser Charles Bastos Rodriguez Denise Demirel Simone Fischer-HuebnerThomas Gross Thomas Langer Mathieu des Noes Henrich C Poehls Boris Rozenberg andDaniel Slamanig Towards a New Paradigm for Privacy and Security in Cloud Services 2015
6 Eric OrsquoShaughnessy Jenny Heeter Chang Liu and Erin Nobler Status and Trends in the USVoluntary Green Power Market Technical Report NRELTP-6A20-65252 National RenewableEnergy Laboratory 2015
References 11
7 Eric Schmidt The Internet Will Disappear World Economic Forum via CNBC TechBet onlinevideo httpswwwyoutubecomwatchv=Tf49T45GNd0 2015
8 Rene Marcel Schretzmann Jens Struckmeier and Christof Fetzer CloudampHeat Technologiesonline httpswwwcloudandheatcom 20112014
9 Internet Society RFC Editor online httpwwwrfc-editororg 199810 Yevgeniy Sverdlik and Karen Riccio Special Report The Worldrsquos Largest Data Centers online
httpwwwdatacenterknowledgecomspecial-report-the-worlds-largest-data-centers 2010
2Architectural Transformations inDistributed Systems
Keywords
Client-Server (C-S) bull Peer-to-Peer (P2P) bull Central Database (CDB) vs Dis-tributed Database (DDB) bull Transactions
The timeline given in the first chapter embodies the perspective of humans using andbenefiting from services In this chapter we now dive under the hood of this developmentand take a look at the service software implementations with a special focus on basicprinciples of complex distributed services which fulfil the requirements for modern cloudand fog applications Over the last two decades we have been able to observe significantarchitectural changes in distributed systems and networking applications which will bereflected in the text There are also mostly orthogonal shifts towards higher reliabilityefficiency scalability and information security as well as other benefitial non-functionalcharacteristics The chapter covers general software and system architectures discussescluster and cloud systems as well as peer-to-peer topologies along with concrete systemexamples and highlights the topics of performance optimisation and transactions as wellas distributed databases
21 Software Architectures and Communication Patterns
Among the most well-known conventional service architectures for software applicationsare the client-server model and the n-tier model In the client-server model a clientconnects to a server to exchange messages with it in order to achieve a certain goal Inthe n-tier model multiple client-server connections exist in a chain Let us consider anintegrated example
copy Springer Fachmedien Wiesbaden GmbH 2017A Luntovskyy J Spillner Architectural Transformations in Network Services andDistributed Systems DOI 101007978-3-658-14842-3_2
13
14 2 Architectural Transformations in Distributed Systems
Fig 21 Example system e-commerce (Source [5])
Example 21 A distributed software application for e-commerce has frequently a rathercomplex hierarchical structure called n-tier which is created with the aim of performanceoptimisation and includes programmatic interfaces linked with network protocols Anexample of a system for e-commerce is depicted in Fig 21 The application 1 for apurchaser (client) interacts with the virtual shop ie application 2 (online shop) via aweb server with the attached application server which provides data preprocessing forpurchase orders The application server for the purchase order preprocessing is connectedto the next two application servers One of them is aimed at store management withmaintainance of store tables the other one at administration of customer data Theapplication 3 supports the communication of the online shop with the suppliers via adedicated communication channel which is connected to an application server as well asthe supplier database Communication between the applications 2 and 3 ie online shop-to-suppliers is performed with use of a corresponding channel provided by the platformThus we see the advancement of typical application architectures to distributed systemswith client-server and n-tier architectures [5 7 8]
As it was shown in [5 8] multi-tier architectures nowadays are widely deployed indistributed applications
bull 3-tier structure is more complex leading to higher scalability preferred for complexapplications
bull 2-tier two-tier structure (user interface and host) is simpler but less flexible (Fig 22)
21 Software Architectures and Communication Patterns 15
Fig 22 Architectures client-server n-tier [8]
Software services Applications or software components which offer service interfacesbeyond their own scope are called software services A typical three-way distinction helpsin distinguishing between services The first kind of service interaction happens betweenlocal service interfaces within a programming language and a corresponding runtimeframework (eg Open Services Gateway Initiative (OSGi) services for Java and othercomponent frameworks) The second kind happens over uniform service interfaces acrossprogramming languages with network transparency (eg web services in service-orientedarchitectures) The third kind happens over non-uniform protocols without obvious siblingor parent protocols and with certain requirements on the topology or infrastructure (egDomain Name System (DNS) Network Time Protocol (NTP) Universal Plug andPlay (UPnP))
Service-oriented architectures have become increasingly popular due to their character-istics They offer a uniform and well-defined interface with the description uniformlycaptured in a machine-processable service description document and accept uniformprotocols with service-specific content Therefore many n-tier applications are nowadaysimplemented within service-oriented systems More recently service designers use thenotion of stateless micro-services which can be replicated easily with coordination througha group communication system What is common to all service-oriented architectures isthe strong reliance on a directory of services called registry through which new servicescan be discovered Sometimes a service broker is available on top of the registry so thatbrokering auctioning and negotiation between service providers and consumers can be
16 2 Architectural Transformations in Distributed Systems
automated in a marketplace style This functionality is important when considering theuser-defined selection of power and computing services covered in the previous chapter
Remote methods and message exchange The interaction between clients and servicesoften follows the request-reply pattern where the client sends a request message blocksto wait for an answer and receives a response message This message exchange styleis similar to local method invocations in programming languages and is therefore alsoknown as remote method invocation Related to this are remote method calls withoutresponse message Complementary to service-oriented architectures there are message-oriented architectures in which software components subscribe to messages of a certaintype arriving from a source to a specific destination or as broadcast message to anydestination In such architectures messages are supposed to traverse message brokerswhich apply filters and transformations An Enterprise Service Bus (ESB) is such abroker which combines service-oriented and message-oriented architectures and facilitatesthe connection of any client to any service with message format adapters
Figure 23 shows a combined service-orientedmessage-oriented architecture Such anabstract architecture will be the basis of many of the systems presented in this chapterwith customisations and refinements whenever necessary
Fig 23 Architectures service-oriented and message-oriented
22 Distributed Service Systems Clustering Grids and Clouds 17
22 Distributed Service Systems Clustering Grids and Clouds
Clusters Significant new features are provided via the clustering architecture in whicheach service is made available in multiple instances (Fig 24) Let us compare it withthe representations which are considered in Figs 21 and 22 The clustering architectureenables the optimisation of the Quality of Service (QoS) for a distributed applicationcaused via functionality replication between multiple servers The functionality forprocessing (application logics) as well as for data persistence is provided via multipleservers simultaneously or parallelised Aimed at replication a preliminary analysis of dataconsistency is required The replication of the functionality optimises the following clus-tering features load distribution fault tolerance behaviour and parallelism in processing(refer to Fig 24)
Server replication in the cluster architecture is characterised via significant gain inthe processing time as pro-argument but also via increasing complexity as con-argumentdue to the conflict management and synchronisation necessity [7] Qualitatively otheropportunities are established by modern architectures of distributed applications forexample applications hosted online or in the clouds (Fig 25)
Fig 24 MPI ndash Message Passing Interface RAID ndash Redundant Array of Independent Disks SANndash Storage Area Network NAS ndash Network Attached Storage Architectures clustering [3 5 8]
18 2 Architectural Transformations in Distributed Systems
Fig 25 Architectures IoS grids and clouds
Clouds The clouds as architectural type provide the deployment and use of ldquocomputingpowerrdquo in a similar manner as by delivering of water or electric current in modern supplynetworks (in so-called ldquoutility gridsrdquo) transparent operation in a ldquocloudrdquo is enabled andpossible The important advantages of the architecture are as follows
bull Sometime the organisations possess insufficient resources for data backup and compu-tational intensive problems then infrastructure outsourcing
bull Aggregation of computing resources of multiple organisations done by the reliable andfavorable providers
bull Companies and authorities obtain a so-called ldquoon-demandrdquo resource access as an idealsolution for fluctuating needs
bull The savings in processing time and hardware costs outweigh the definitely noticeablegrowth in the coordination and synchronisation complexity
The disadvantages are as follows Cloud computing fosters heterogeneity vendor lock-in through attraction by vendor-specific cloud services as well as an unclearness ofdata security protection aspects when the data processing crosses organisational or evenjuridical boundaries
There is no single definition of what a cloud system is A commonly used definitionis given by National Institute of Standards and Technology USA (NIST) 2011 ldquoCloudComputing is a model for enabling ubiquitous convenient on-demand network access
23 Architectures Peer-to-Peer 19
to a shared pool of configurable computing resources (eg networks servers storageapplications and services) that can be rapidly provisioned and released with minimalmanagement effort or service provider interaction This cloud model is composed of fiveessential characteristics three service models and four deployment modelsrdquo [4]
There are scientific community and voluntary cloud systems accessible to everybodyat no or low cost but also no strict service-level guarantees Examples include Guifi andOwncloud instances On the other hand there are commercial cloud providers who offerrapid provisioning and elasticity of resources at large scale Examples include AmazonEC2 IBM Softlayer and Bluemix T-Systems Enterprise Cloud and the Google CloudPlatform
Grids One of the most important parts of cloud technology are the grids The termldquoGRID (Global Resource Information Database)rdquo was founded in 1985 as part of a UNOprogram for environmental protection on the other hand ldquoGRID=SUPPLY NETWORKrdquoIn this way grid is the traditional technology in comparison to clouds Grid Computingis a form of distributed computing in which a ldquovirtual supercomputerrdquo from a cluster ofloosely coupled computers is generated The grids were developed with the aim to solvethe computational-intensive scientific and in particular logical-mathematical problemsThe following types of grids are distinguished regularily [5]
bull Computer grid a combination of computing power and allows the access to distributedresources
bull Data grid enables sharing of data contained in the requests of one or more DDBbull Service grid represents diversity of components all of them belong to different
resource providers and are provided as a utilitybull Application grid provides improved load balancing and utilisation of grid provider as
well as wide spectrum of functions via a cross-organisational sharing of resourcesbull Resource grid has to be defined via a rolesrsquo model deployment ie the roles
between grid users -providers and the resource providers which have to be clearlydifferentiated [1]
A basic understanding of clusters grids and clouds should now be achieved The twoformer system types will be elaborated on in much greater detail in the next chapter andthe latter one in the subsequent one
23 Architectures Peer-to-Peer
In the contemporary fixed-mobile converged networks with almost-always-on connectivityover Local Area Network (LAN) Wireless Local Area Network (WLAN) as well asWireless Personal Area Network (WPAN) routes the role of peer-to-peer architecture
20 2 Architectural Transformations in Distributed Systems
Fig 26 Examples of peer-to-peer systems
with equal partners (P2P) has been significantly increased Let us compare the architecturewith the already presented one titled ldquoclient-serverrdquo (C-S) P2P offers
bull Direct communication between the equal partners ie the peersbull Practically no centralisation within the server part only as an option in the combined
(hierarchical) structures involving peers + C-Sbull The peers are simultaneously the service providers as well the users or consumers of
the servicebull A distributed discovery mechanism for service providers as peers is required
As it is depicted in the representation (Fig 26) the peer-to-peer model (P2P) enhancesthe client-server model (C-S) towards a multi-participant fully-connected bi-directionalone In the C-S model a server provides a service and a client uses this service In P2Pnetworks this role for distribution is without meaning Each participant is a peer becausethey can use a service as well as offer the service by themselves The basic requirementto the architectures with equal partners providing P2P communication is the creation of aspecial mechanism to search for service provider peers The following types A B C andD of the P2P-systems are distinguished [6]
bull Type A called centralised P2P model which uses the server core for co-ordination andsearch Example Napster
bull Type B called pure P2P model provides no centralised co-ordination ExampleGnutella
bull Type C called hybrid P2P model The dynamic centre contains the entities some peersact but as the coordinators The examples are Gnutella2 BitTorrent Skype
23 Architectures Peer-to-Peer 21
bull Type D called distributed P2P model with the DHT Distributed Hash Table Thetable manages the access IDs ie the keys are placed on a carouselcircle The modeluses overlapping of fixed connections (Fixed Connection Overlay) The system issimilar to the well-known routing protocols for distributed (RIP) or hierarchical routing(OSPF+BGP) The examples are as follows Chord CAN Pastry Tapestry
Figure 27 contains a visual representation of all four types of P2P systemsThere is a trade-off in enabling a P2P architecture for a distributed application On the
one hand it puts an obligation on all participants to offer a share of their resources to otherparticipants as only through a fair distribution such a network will work well As withall service interfaces open to the world over a network there is a risk of being attackedthrough the interface On the other hand once a sudden allocation of resources is neededthe scalability of such a network especially on a global level with high availability andresilience is very high and cannot easily be reached with other architectures Thereforeespecially for applications which involve humans including all personal communicationpersonal information management and personal cloud activities P2P architectures areeffective
Example 22 Modern mobile client platforms provide many attractive mobile applica-tions and transmission services in addition to the standard voice SMS MMS and E-mailA number of these services include for example the popular Google Suggest GoogleTranslate Google Maps cloud services Amazon AWSEC2 social networks Facebook
Fig 27 Types of P2P architectures [5 8]
22 2 Architectural Transformations in Distributed Systems
Fig 28 Skype network structures and diverse clients
Twitter Xing video hosting service YouTube as well as multiple VoIP services like Skypeand Viber The designated service Skype is now a leader by a wide margin among the manyVoIP services The service is generally available for free and supports the following built-in services VoIP televideoconferencing instant messaging transfer of files images andscreenshots Surely Skype is the worthy rival to many VoIP services with use of commonprotocols like SIPRTP and SIPUDP But a lot of them are only available with costlycommercial plans In contrast Skype is aimed at the private sector and offers the followingfeatures (Fig 28)
bull wide availability despite of proprietary (not published) protocol (cp SIPRTP)bull optimised hybrid architecture P2P + C-S with central servers core run by Skypersquos parent
company Microsoftbull data compression and proven security via AES with 256-bit key RSA with 2048-bit
key as well as X509 PKIbull IPv6 as well as IPv4-based and transparent for NAT therefore suitable for home usersbull data compression with the codecs SVOPC (16 kHz) AMR-WB (16 kHz) G729 (8
kHz) G711 since 2009 an own audio codec SILK is usedbull compatibility to conventional telephony gateways to conventional telephone networks
(PSTN ISDN GSM)bull integration with SIP-based VoIP
24 Performance Optimisation 23
Herewith a short but worthy history of Skype service The Skype core software wasdeveloped by Ahti Heinla Priit Kasesalu and Jaan Tallinn (Estonia 2003) The companyfounded by Niklas Zennstroumlm and Janus Friis in 2003 in Luxembourg but since 2005 wasowned by eBay and in 2011 was transferred to Microsoft
The following archictectural transformations are to be watched via its history Theoriginal Skype network (2003ndash2010) was characterised via primary P2P organisation likea lot of multimedia sharing systems with so called nodes and super-nodes The systemoffered voluntary reallocation on own private computers but suffered due to very busy(overloaded) super-nodes The private client machines required especially IP without NATwith direct addresses As an upshot increasing criticism from private users as well asnumerous failures due to overloading of peers were noted After the takeover by Microsoft(2011) the following architectural changes were applied
bull cleaned structures steady Skype network restructuration since 2012bull Skype network was transferred from the client computers to its own Linux servers (ie
partially from P2P to C-S)bull currently P2P with a centralised C-S constructionbull server clusters are placed at secure data centers (PaaS delivered through clouds)bull enhanced security of Skype servers is guaranteedbull software development under Skype is available Skype API allows use of the Skype
network for delivery of messages and call management
24 Performance Optimisation
Methods for performance optimisation As you have seen from the introductionmodern distributed network systems are used in the areas of Business-to-Business(B2B) Computer-Aided Design (CAD) Grids and Cloud Computing They aredeveloped to solve complex mathematical tasks actual problems of modern pharmacologyto simulate physical phenomena and in genetics to administrate and manage task supportThese systems process and transmit via networks significant amounts of structureddocuments and multimedia data which for extreme volumes has recently gained the termBig Data In general the following performance optimisation methods [56] can be appliedwithin the classical C-S as well as new architectures of distributed systems like clusters andclouds (Fig 29)
bull Cachingbull Replicationbull Parallelisation
24 2 Architectural Transformations in Distributed Systems
Fig 29 Performance optimisation [5]
Frequently used addresses and names should be cached Caching can be deployed onthe site of the server as well on the site of the client or is present within the networkinfrastructure typically outside of the scope of application deployment The client-sidecaching is often very efficient Another method is the data and services redundancy viareplication Server replication can be efficiently used for load balancing in highly-availablemulti-server systems as well as to provide a certain level of fault tolerance through failovermechanisms Parallel processing within a server application follows frequently under useof multiple execution processes or threads Process parallelisation and multi-threadingmode provide significant performance increase All three methods are quite generic andcan be found in most scalable applications to overcome performance bottlenecks
The following empiric rules are known which are required when optimising perfor-mance in distributed systems particularly in systems of the type C-S [5]
bull The CPU speed is often more important than the network performance and can becomea bottle-neck
bull Reducing delays in processing of application protocols (software overhead) throughaggregation packets has a measurable effect
bull Minimising context changes between the processes (in multi-threading) makes applica-tions faster
24 Performance Optimisation 25
bull Minimising the backup and copy processes within the system for example due to useof shared memory devices
bull The important requirements to increase the data rate are not so critical as delay eveninsignificant
bull System overload is easier to prevent than to overcomebull Preventing timeouts and pauses within the system reduces unnecessary slowdowns
Threads A thread is a so called ldquolight-weightrdquo independent subprocess running inparallel to other (sub)processes which can be considered as a part of a complexapplication The thread is operated without or with minimal context sharing to other (sub-)processes and threads but with its own program counter and existing stack (Fig 210)
Typically the application processes that are performed in a certain Operating System(OS) (as programming environment) are ldquoheavy-weightrdquo due to the large amount of thecontext (process parameters) to be transferred
A well-known example for such ldquoheavy-weightrdquo processes are the ones that areperformed in the operating system UNIX and derived systems such as Linux as usedin GNULinux and Android among others BSD and Darwin the kernel of Mac OS X Toprovide some additional flexibility and parallelism within them each complex process isdivided into so-called ldquolight-weightrdquo sub-processes that are specifically called threads Athread is de-facto a bearer of certain activity within an OS or programming environment
Fig 210 Px or P123 ndash complex processes or applications Txy ndash thread a ldquolight-weightrdquoparallelised sub-process without dependencies but with own program counter and stack Anapplication as a combined process Px with several threads Txy
26 2 Architectural Transformations in Distributed Systems
This action is performed via a set of consecutive operations and is characterisedby a minimal context consisting of only stack and registers In practice most of thecomplex applications and system processes are suitable for implementation in the formof parallelised threads Each of these flexibly distributed ldquoheavy-weightrdquo processes has atleast one initial thread as ldquolight-weightrdquo sub-process All such threads which are merely apart of some greater processes are used within the same common address space as otherresources of the complex process
Example 23 There is the following simple example With the mentioned methods a wordprocessor application (eg MS Word) can be divided onto several parallel threads whichcarry out over one and the same data (text) within a file a set of various operations forinstance (1) text splitting (2) text formatting and (3) spell checking
Example 24 In addition the applications that performs a large number of independentasynchronous requests (ie database applications server-side web applications) alloweffective implementation with deployment via several parallel threads as multithreadedappplications Generally there are the following two types of threads
1 The user-level threads which are realised from the scope of view of an applicationprocess via a programmer
2 The kernel-level threads or kernel threads which are used for representation ofan OS for example MS Windows and its programming environment with the aim toperform them at a certain processor
25 Distributed Transactions
Using transactions several actions can be combined with the aim to form an indivisibleexecution unit T
T D A1 A2 A3 (21)
These can be also called atomic (trans-)actions ie with use of the slogan ldquocompleteor nothingrdquo An example of a transaction monitor is given in Fig 211 The monitorcoordinates the booking workflow between C and S1 S2 The finalising phase is veryimportant and has to be involved with the aim to support the consistency of data
The 2PC protocol must be used to ensure consistency in this way ldquoCommitrdquo inthis context means consensus agreed to meet requirements or to depute The diagramdepicts a successfully executed transaction with a reliable storage device which guaranteespersistency such as a disk external storage medium or reliable storage service (Fig 212)
Beyond consistency and persistency the transactions have to satisfy the so-calledAtomicity Consistency Isolation Durability (ACID) criteria The deployment of the
25 Distributed Transactions 27
Fig 211 Transaction monitor
Fig 212 Sequence diagram for the 2PC protocol
28 2 Architectural Transformations in Distributed Systems
Fig 213 Distributed transactions deployment of 2PC [5]
distributed transactions is also based on considering common methods for performanceoptimisation These criteria called ACID describe the desirable properties of all types ofthe transactions The transactions have to ensure the ACID criteria
bull Atomicity Either full execution or completely without effectbull Consistency Transformation only between consistent statesbull Isolation No overlap of parallel transaction executionsbull Durability Survival of system failures
An example of the use of 2PC is depicted in Fig 213 The example illustrates providingatomic actions under the slogan ldquocomplete or nothingrdquo The appropriate realisation with2PC ensures atomicity as one of the ACID criteria The user has to be provided via a travelagency two flights (eg with Lufthansa and United) as well as with a rented car at thedestination site If the booking is impossible the consistent rollback cancels all actionswithout financial disadvantages for the user or the agent
The protocol uses the following messages C-Refuse from the participants if one ormore rejections then send C-Rollback if necessary then repeat The realisation can bedone via ODBC or JDBC (ObjectJava Database Connectivity) when run in a databasecontext Performance increases are available with the deployment of parallel transactionsobeying to the isolation criteria
bull Optimisation by redundant reservation of server processes (separate servers)bull Parallel execution via multi-threading
25 Distributed Transactions 29
bull Replication of servers (replication)bull Heuristic load balancing and reliability
The appropriate example is depicted below (Fig 214) In the offered parallel transactioninstead of one three servers and a replicated DB are used
Figure 215 depicts a nested transaction in a travel booking scenario It starts witha successful booking of an appropriate room but then mandates a rebooking activity of
Fig 214 Parallel transaction instead of one three servers and a replicated DB are used [5]
Fig 215 MSP ndash main synchronisation point coordinated by careful Commit AffSP ndash affiliatedsynchronisation point the action Activity allows partial rollback FSP ndash final synchronisation pointterminates the instances Nested transaction involving multiple independent partners in a travelscenario
30 2 Architectural Transformations in Distributed Systems
two necessary flights to Incheon International Airport in Seoul from Dresden (DRS) viaFrankfurt-am-Main (FRA) airport or Munich (MUC) airport due to no longer availableseats Due to a changed meeting request the travel is finally substituted via another tripfrom Dresden central station to Zurich with the night train (CNL) with a successful finali-sation (FSP final synchronisation point) the instances are terminated To ensure the ACIDcriteria within the nested transaction the MSP (main primary synchronisation point)coordinated by careful commit as well as AffSP (affiliated secondary synchronisationpoint) are used The action Activity allows a partial rollback
Thus depending on the application scenario and requirements transactions may bedistributed parallel and nested
26 Distributed Databases
Motivation for DDB The deployment of the distributed DB takes into account the abovementioned common methods for performance optimisation Let us give the definition ofa DDB We consider it in contrast to the centralised DB (CDB) A distributed database(DDB) possesses the following features (Fig 216)
1 DDB forms a logical unit2 DDB is physically stored on separately located computers (homogeneous or heteroge-
neous)3 DDB requires a communication network4 DDB has no shared memory5 DDB appears to users and applications as a CDB
But it is important to note that not each distributed system needs a DDB A central (globalDB) can be also used as an efficient solution for instance in an n-tier-architecture Ineach case it has to be individually decided which type of DB is the most appropriatewhile taking into account the performance optimisation methods There are the followingarguments for comparison of both kinds (CDB vs DDB) Which arguments are thesatisfying motivations for distributed databases which advantages are available
bull higher performance and faster accessbull higher availabilitybull more security in the sense of confidentialitybull reduced communication costsbull faster query processing in the Structured Query Language (SQL)bull increased extensibility and scalabilitybull adaptive scalability by fluctuations within the user number node quantity quantity of
the records of rows within the DDB number of the queries to process etc is offered
26 Distributed Databases 31
Fig 216 Decision making CDB vs DDB
To the disadvantages list of DDB the following restrictions can be assigned as follows
bull increasing complexity of the systembull overhead by commit operationsbull data integrity problemsbull increased memory requirements
Up-to-date solutions for databases nowadays generally possess the 3-tier-architecture TheCDB consists of
bull internal schema (logical layer) which determines the physical structure of the data onthe disks
bull external views which define the data visualisationbull conceptual layer as an interface between internal and external (Fig 217)
Decomposition methods A characteristic unique to DDB is that specifically the concep-tual scheme is divided into a global and many local schemes (Fig 218) With the goal ofdecomposition of the conceptual scheme of a DDB into many local schemes the followingmethods are available replication or fragmentation as follows
bull by replications (full copies regular backup)
32 2 Architectural Transformations in Distributed Systems
Fig 217 Classical DB three layers
bull horizontal (line-wise) decomposition (fragmentation by tuples)bull vertical (column-wise) decomposition (fragmentation by attribute subsets)
Generally the description of the mentioned access levels to the DDB via the followingspecial languages can be used
bull DDL ndash Data Definition Languagebull DML ndash Data Management Languagebull QL ndash Query Languagebull DSDL ndash Data Storage Definition Language
DDB fragmentation Fragmentation of DDB within distributed applications can offer thefollowing advantages
bull efficiency data are located where they are really neededbull local optimisationbull increased availability and security better DB view demarcationbull no data losses simple recovery of DDB is available via ldquounionsrdquo and ldquojoinsrdquo from E
Codd
26 Distributed Databases 33
Fig 218 Layered architecture within DDB
As disadvantage acts the risk of inconsistency by access runtimesAn example of the DB fragmentation is given in Fig 219 The relation table titled
ldquoEmployees by departmentsrdquo is a CDB which is situated locally (referring to (a)) Withthe aim of performance optimisation this CDB is decomposed via a fragmentation methodRefer to the cases (b) and (c) for horizontal and vertical decomposition correspondently
bull Horizontal (line-wise) decomposition with use of fragmentation by tuplesbull Vertical (column-wise) decomposition with use of fragmentation by attribute subsets
Replication of DDB The advantages of DDB replication are as follows
bull increased availabilitybull reliability easier backupbull increased access performance
A resulting problem is that replicas may be out of date when they are accessed while themaster data has just been modified Furthermore more problems occur when attempting tosynchronise the data when changes may occur not just in one master node but in multiple
34 2 Architectural Transformations in Distributed Systems
Fig
21
9(a
)R
elat
ions
tabl
eldquoe
mpl
oyee
sby
depa
rtm
ents
rdquo(l
ocal
DB
)(b
)H
oriz
onta
l(lin
e-w
ise)
deco
mpo
sitio
n(f
ragm
enta
tion
bytu
ples
)(c
)Ve
rtic
al(c
olum
n-w
ise)
deco
mpo
sitio
n(f
ragm
enta
tion
byat
trib
ute
subs
ets)
DD
Bde
com
posi
tion
via
frag
men
tatio
n
26 Distributed Databases 35
nodes concurrently This multi-master replication compared to master-slave is howevermuch more scalable for write operations while the scalability for read operations remainsunchanged
Therefore when planning the deployment of a distributed database the followingreplication-related questions need to be answered carefully
bull How many copies are required in order to achieve either a high scalability or a highavailability
bull Where do the copies have to be storedbull What will be the dominant access pattern read or write access
Efficient updates in DDB are possible
bull Requirementndash replication of DDBndash full copiesndash regular (automated) backup
bull UPDATE mechanismsndash Primary copy-scheme (asynchronous method)ndash Majority consensus scheme (synchronous method)ndash locking tablesndash logic time stamps
bull Requests and concurrencyparallelismndash local and global transactionsndash requests in standardised SQL dialectndash actual data structure for users or applications is unknown or not definedndash communication overhead times are significantly higher in comparison with comput-
ing timebull As a solution
ndash local pre-processing (so much as possible)ndash exchange with partial results (so called ldquosemijoinsrdquo)ndash ACID and 2PC-protocol
bull Steps
1 decomposition of the requests into simple partial requests2 locating the required data decision which copy is used transforming into the partial
requests depending on a network node3 optimisation of the global request (order processing)
A 2PC example for DDB is given in Fig 220 The example is about the coordinationbetween the parts in four geographically separated cities eg Berlin (DDB0) Dresden
36 2 Architectural Transformations in Distributed Systems
Fig 220 2PC example for DDB
Cologne and Hamburg (DDB123) For the consistency of SQL requests from thecoordinator or the main part DDB0 the messages Commit 123 or Rollback 123 areused
The following variants of commitment by SQL requests processing are possible withinuse of DDB via the 2PC
bull Succesful variantndash SQL requestndash A local transaction is finalised as OKndash Preparation to COMMITndash Prepare COMMITndash Ready 1 2 3ndash Commit 1 2 3ndash Commit ACK 1 2 3
bull A failure variant the replication 3 offers no commitndash SQL requestndash A local transaction is finalised as OKndash Preparation to COMMIT
26 Distributed Databases 37
ndash Prepare COMMITndash Ready 1 2 Abort 3ndash Rollback 1 2 3ndash Rollback ACK 1 2 3
The following synchronisation (also voting co-ordination) methods within theDDB are available for implementing the instruction ldquoUPDATErdquo for the availablereplicas [8]
bull Primary-Copy-Schema (PCS) (asynchronous)bull Majority-Consensus-Method (MCM) (synchronous)bull Locking tablesbull Logical timestampsbull Protocols like two-phase-commit and two-phase-lock (2PC Two-Phase Lock
(2PL))
The asynchronous PCS is a process for the synchronisation [2] of replicated data In thismethod the change is performed only on the primary copy and then synchronised withthe replica The primary copy will prevail The advantage of the method is that if thereare several changes they can be bundled to be synchronised with the other copies Thedisadvantage is that the method does not ensure a stable consistency for the distributedcopies [2]
This is in contrast to the MCM which is a synchronous method The main principle forMCM is as follows The update on a copy will be carried out only if the correspondingtransaction is able to win a majority of copies (eg is relevant to lock) In principle thereare multiple possible MCM variants The MCM differ from each other with the followingaspects First whether all copies of this voting can be treated equally (unweighted voting)or not (weighted voting) and second whether the number of the votes which are requiredfor reaching the majority is fixed (static quorum) or this number can be computed only atrun-time (dynamic quorum)
Note For the read access (read quorum) and for the write access (write quorum) adifferent number of votes have to be defined [2]
Among further synchronisation methods the locking tables logical timestamps as wellas 2PC2PL or their combinations should be briefly mentioned These methods (usuallycombined) are distinguished by the following characteristics
bull Locking tables ie blocking of unwanted changes in certain replicates (like PCS +MCM)
bull Logical timestamps ie monitoring by the timestamps then like PCS
38 2 Architectural Transformations in Distributed Systems
27 System Examples Google Spanner a Global DDB
Some of the more sophisticated DDB systems are offered by the commercial serviceprovider Google Among them are
bull Google Bigtable (2008)bull Google MegaStore (2011)bull Google Spanner (2012)
There are also the further known relational and non-relational DDB from commercial ven-dors (IBM Sybase Oracle Microsoft) and open source projects (Cassandra CouchbasePostgres-XC Postgres-R) Many of the following explanations also apply to these systemson an abstract level
Spanner was developed to resolve the disadvantages of Googlersquos Bigtable and MegaS-tore [3]DB Bigtable (2008)
bull difficult deployment for complex and self-evolving schemasbull no strict consistency guarantees for geo-replicated sites (partitions)
DB MegaStore (2011)
bull synchronous replication and semi-relational data modelbull full ACID semantics in the partitions but only small consistency guarantees on
partitionsbull low write throughput
A typical world-wide deployment scenario for Spanner is shown in Fig 221 On eachcontinent a number of data centres are running instances of the database This guaranteesa low-latency access from nearby users and avoids overloading a single instance
The internal architecture of a distributed Spanner installation is explained in Fig 222Each site is called a zone and coordinated by a zone master All zone masters are in turncoordinated by a universe master Furthermore location proxies take the requests fromdatabase clients and forward them flexibly to span servers
The following terms and quantities are of relevance when looking at the architecture
bull Universe the overall deployment areabull Zones deployment area for servers in one site physically isolated units placement and
distribution driverbull 1 Universe masterbull 1 Zonemasterbull 1000 Spanservers
27 System Examples Google Spanner a Global DDB 39
Fig 221 Deployment scenario online social networks
Fig 222 Spanner architecture [3]
For the realisation of Spanner a specific software stack modelled around the Paxosalgorithm has been designed Fig 223 offers a look inside the stack
Building on Spanner there is the newer system Google F1 SQL called the ldquoFault-Tolerant Distributed RDBMSrdquo As a replacement for basic relational systems like MySQLor PostgreSQL it offers the following features
bull NewSQL platformbull Each Span-Server is responsible for 100 up to 1000 Tablet instancesbull Data and log files are stored on Colossus a successor of the Google File System
40 2 Architectural Transformations in Distributed Systems
Fig 223 Spanner software stack [3]
bull Paxos is used for commits (consensus) for all participants a common value matchesbull Paxos is used for consistent replicationsbull A Transaction Manager for distributed transaction support 2PCbull True Time Architecture
Paxos is a traditional algorithm named after the Greek isle of Paxos next to Corfualthough originally by the author of the algorithm erroneously placed into the AegeanSea It works as follows
bull Server can act simultaneously as proposer acceptor and learnerbull During normal operation the leader receives a clientrsquos command assigns it a new
command number i runs i-th instance of the consensus algorithmbull Paxos group all machines involved in an instance of Paxosbull Within Paxos the group leader may fail and may need re-election but the safety
properties are always guaranteed
The workflow of Paxos is shown in Fig 224
Apart from implementing Paxos Spanner offers the following architectural properties
bull scalable multi-versioned global-distributed synchronously replicated databasendash distributed transactions (with 2PCACID)ndash SQL-driven schematic tablesndash but semi-relational data model
27 System Examples Google Spanner a Global DDB 41
Fig 224 Paxos algorithm
ndash reconfiguration of replications is very fine-grainedndash dynamic reconfiguration per application
bull Applications can define the parameters and constraintsndash such as the number location and distance of replications
bull Dynamic data migrationndash data can be transparent moved at a global level even during operationndash consistent read and write access
bull Aims and focusndash management of cross-replication of datadata centersndash global consistent writes via Google Spanner
bull Deployment examples up-to-date productsndash Google Ad Data (Advertisers)
bull 50 Paxos groups 2500 directories read- and write access of 4 KBytendash commit within ca 5 msndash latency generally under 9 ms
bull True Time several thousands Span servers at a distance of max 2200 km (withoutlatencies due to distance)ndash 90 no deviationndash 9 deviation up to 2 msndash 1 deviation up to 10 ms (still far too much)
This architecture allows for creating complex applications Picking up the previousexample of a social network installation again a Spanner-based application may look likeshown in Fig 225
To synchronise the distributed database Spanner a protocol of real time is used calledTrue Time (Fig 226) In order to implement the controlled access not only time stamps
42 2 Architectural Transformations in Distributed Systems
Fig 225 (a) Single machine (b) Multiple machines Sample application of DDB with Spanner [3]
Fig 226 True Time message exchange
are used but full time intervals The replica synchronisation is performed every 30 s Tocorrect the time GPS and atomic clock usage is foreseen The quasi-parallelism of theaccess is provided for two access modes
bull The ldquoread-onlyrdquo access proceeds in the ldquosnapshotrdquo modebull The ldquoread-writerdquo access proceeds via the 2PC and 2PL protocols [3]
28 Conclusions 43
Table 21 True Time methods True Time API method Time output
TTNow() TTinterval [earliest latest]
Boolean TTAfter(t) True if t has definitely passed
Boolean TTBefore(t) True if t has definitely not arrived
For programmers True Time offers three convenient methods to deal with relative andcausal times They are explained in Table 21
Therefore the examined DDB Spanner system possesses the following metrics andperformance parameters [3]
bull 50 Paxos groups and 2500 access directories are createdbull The read and write access proceeds for the data portions (called chunks) with minimum
size of 4 Kbytesbull A middle commit can be reached within approx 5 msbull The summarised request latency is no more than 9 ms
The True Time protocol provides the ability to use thousands of so-called Span serverslocated at a considerable distance from each other They work without significant delaydespite considerable distance to a maximum of 2200 km The following access statisticshave been observed
bull In 90 of the cases there is no deviationbull In 9 of the cases the deviation reaches up to 2 msbull Nevertheless only in 1 of the cases the deviation obtains a significant latency of
10 ms or more
Further system examples for DDB are associated with the databases which belong to well-known manufacturers like IBM Sybase Oracle or Microsoft
28 Conclusions
The architectural solutions for modern distributed systems and networking applicationshave been subject to significant changes in recent years Modern architectural transforma-tions contribute to the development of new attractive for users (mobile) services searchengines content management systems custom video hosting services cloud servicesVoIP tools social networks There is no possibility to specify a complete list Dependingon the needs of the application and ultimately its users a concrete software architectureand communication pattern (C-S P2P) needs to be chosen Assuming performancematters performance optimisation methods should be evaluated and applied For higher
44 2 Architectural Transformations in Distributed Systems
reliability data processing tasks should run in transactions Distributed databases suchas Spanner are already optimised for global high-performance deployments and thereforefree the application engineer from labor-intensive and error-prone custom methods
References
1 C Baun M Kunze J Nimis and S Tai Cloud computing ndash Web-based dynamic IT-ServicesSpringer-Verlag 2010 in German
2 P Dadam Verteilte Datenbanken und ClientServer-Systeme online httpwwwinformatikuni-ulmdedbispapersvdb-buchvdb99_09pdf 1999
3 J C Corbett et al Global Distributed Database Google Spanner Berlinbuzzwords 20124 P Mell and T Grance The NIST definition of cloud computing whitepaper NIST Special
Publication 800-145 September 20115 Alexander Schill and Thomas Springer Verteilte Systeme - Grundlagen und Basistechnologien
Springer-Verlag second edition 2012 433 p in German6 R Steinmetz and K Wehrle Peer-to-Peer Systems and Applications Springer 20057 Andrew S Tanenbaum and Maarten Van Steen DISTRIBUTED SYSTEMS Principles and
Paradigms Pearson 2013 633 p8 Andrew S Tanenbaum and David J Wetherall Computernetzwerke Pearson Studium fifth
edition 2012 1040 p in German
3Evolution of Clustering and Parallel Computing
Keywords
Clusters bull Grids bull Performance parameters bull High-Performance Computing(HPC) bull Speedup models bull Amdahl model bull Barsis-Gustafson model bull Karp-Flattmetric bull Berkeley Open Infrastructure for Network Computing (BOINC)
Demarcation between parallel and distributed computing clusters and grids Theparallel execution of code within applications is a standard feature for higher performanceresponsiveness or both Parallel code the building block for parallel computing isachieved by multiple processes multiple threads co-routines and similar programmingtechniques Typically parallel code is assisted by hardware such as multiple processorsper node or multiple processor cores per processor (virtual processors) and otherwise bythe operating systemrsquos process scheduler (pseudo-parallelism)
The effects of parallelism on the execution time of an application are shown in Fig 31When the hardware support extends to multiple connected nodes with appropri-
ate messaging techniques the extended paradigm of distributed parallel computing isachieved The connected set of nodes is then often called a cluster Of course applicationscan also be parallelised without hardware support but there will be only gains whenthe computing resources (processor memory disk or network) are not yet exhaustedThe terms high-performance computing (HPC) and high-throughput computing (HTC)express respectively focus on a subset of these resources and attempt to maximise theirusage This claim is not essential to distributed computing per se
Another perspective at parallel code execution and clustered nodes is the approach ofhow to use the system When a large set of nodes is connected and offers the submissionand computation of jobs from a bag of tasks the resulting system is called a gridIn recent times with the on-demand provisioning and elastic scaling of resources as
copy Springer Fachmedien Wiesbaden GmbH 2017A Luntovskyy J Spillner Architectural Transformations in Network Services andDistributed Systems DOI 101007978-3-658-14842-3_3
45
46 3 Evolution of Clustering and Parallel Computing
Fig 31 Effects on parallelism (a) no parallelisation (b) hardware parallelisation (c) pseudo-parallelisation by a scheduler
well as usage-based billing of computing resources (utility computing) the dominatingterm instead of grid is rather cloud leading to the more recent paradigm of cloudcomputing although volunteers around the world still connect their personal computersin desktop-based grids called volunteer computing and meshes when the focus is moreon networking capabilities [26]
The foundations to the organisation of the parallel computing process based on gridsclusters and clouds are discussed in [4 8] with a practical look on grid and cloudintegration in [7] and additional research trends listed in [17] Education on these topics isdiscussed intensively in [1 10]
While the next section will introduce several counters units and scales to comparethe performance of computing systems one should already be introduced here to give arough sense of comparison between diverse computing architectures The unit of choicehere is Floating-Point Operations Per Second (FLOPS) most often used in the scale ofTFLOPS or 1012 FLOPS
Typically grids differ from clusters by geographical dispersion of and public access toits computers and are characterised via significantly heterogeneous structure In additioneach grid generally uses standardised software components for co-operation and commu-nication (standardised Application Programming Interface (API) libraries middlewareweb services) One of the prominent early examples is the first Metacomputing system bythe University of Illinois [24] On the other hand clusters are centralised and possess ahomogenous structure with powerful CPUsGPUs as well as SANNAS for data storageIncreasing efficiency and reducing heterogeneity is possible with the use of off-the-shelfcomponents open-source operating systems and resource virtualisation (networks pro-cessors memory devices applications) For high-speed data transfer between processorseither Ethernet (1 GBits) or fibre-channel technology (FC eg 16 GBits fiber channels)is used Deployment of powerful clusters as well as loosely coupled and grid-connectedprivate PCs tablets and even smartphones create virtual supercomputers which providea high performance As mentioned one measurement unit for the performance is thenumber of FLOPS Todayrsquos supercomputers achieve multiple TFLOPS or even PFLOPS(Taurus Titan Tianhe-2) These supercomputers can be aimed at parallel solving ofcomputationally-complex math-log cooperative problems More modest cluster systemsexist including the Beowulf design applicable to small-scale installations [1 14] Among
3 Evolution of Clustering and Parallel Computing 47
the international grid systems for parallel computing the BOINC grid [28] is one of themost well-known ones although newer systems such as OurGrid and the European GridInfrastructure (EGI based on federated clouds) still offer functional innovation [5]
Example 31 Many educational institutional and national grids reflect the evolutionalchanges in grids and high-performance computing during all time of its existencefrom appearance until modern trends [18] The Ukrainian National Grid together withURAN (Ukrainian Research Academic Network) and some dedicated projects is a typicalrepresentative of this observation [19 20] It offers two middleware resource types asremote service gLite and ARC Many national research laboratories universities andinstitutes offer concrete service realisations In total 27 ARC services and 2 gLite servicesare provided Among the providers is the Institute for Condensed Matter Physics whichruns an ARC site with 17 compute nodes 3 storage nodes and a coordinator nodein a cluster format This cluster achieves about 11 TFLOPS whereas the overal gridperformance is much higher
Another example is SwiNG the Swiss National Grid Its network consists of thescientific computing centres of 18 higher education institutions and research institutesThe Ukrainian National Grid intends to participate as a member grid in EGI and SwiNGis already a member grid along with more than 30 others EGI in turn intends toevolve jointly with other partners into the European Open Science Cloud for ResearchThis endeavour is built on eight fundamental elements for success among them serviceorientation and interoperability
In general there have been the following essential phases in the development towardstodayrsquos clusters and grids
1 Meta-computing pioneer grid projects like GRID and the Metacomputer based onactive involvement of the technologies from scientific areas to everyday life
2 Convergence with web technologies (eg BOINC) wide-spreading of grids throughinstitutions and volunteers
3 Efforts to solving of wider range of problems secured access interoperability resourcediscovery on the basis of deployment of standardised middleware like OGSA (OpenGrid Services Architecture)
4 Wide-spread acceptance of grid services in the same way as delivering of waterand electricity and then inset of the SOA approach (service-oriented architectures)via standardised web services deployment and workflow composition (WS-BPELBusiness Process Execution Language)
5 Wide-spreading of cloud computing as a model for enabling ubiquitous convenienton-demand network access to a shared pool of configurable computing resources withessential measured services like Everything-as-a-Service (XaaS) and rapid elasticity
48 3 Evolution of Clustering and Parallel Computing
6 Integration of grid services within high-available clouds (mostly PaaS) together withparallel clusters (IaaS) and capable network storages (RAIC Redundant Array ofIndependent Clouds)
7 Development of new energy-efficient grids clusters and cloud services smart gridtechnology with a link to power distribution systems to combine computing on demandwith power on demand
Recent tendencies in the usage of parallel computing for the simulation of technologicaldevices and processes including electron beams and electron guns indicate a rise ofsmall but smart low-energy clusters They are based on multicore CPUs built-in withinregular PCs such as Intel Core i7 Core i4 or AMD FX in the kWh-area or even ononboard microcontrollers like Raspberry Pi Arduino or Intel Edison with only lowWh-consumption
In the remainder of this chapter performance parameters and models will be presentedfollowed by a discussion of trade-offs and a presentation of modern frameworks to manageboth resources and applications in cluster and grid environments The discussion of cloudcomputing and smart grid concepts respectively is then following in the subsequentchapters
31 Clustering and Grids Performance Parameters andBasic Models
Performance parameters Let us first define the most important performance factorsand metrics beyond the already mentioned FLOPS The code execution performanceparameters of modern computers are as follows [23]
bull Number of CPU coresbull Tact (clock) frequency per core f unit 1
s D Hzbull Million Instructions Per Second (MIPS)bull FLOPS as defined above
The system clock signal produced by a crystal oscillator synchronises the operation ofmultiple functional blocks within a CPU The system tact is a periodical function basedon the Peirce function using the negated logical OR operator NOR Some examples ofthe performance of certain CPU models from recent production years are given below(Table 31) It is evident that the tact frequency is no longer the dominant differentiatorbetween CPUs Rather the number of cores enhanced throughput and parallelism and ahigher efficiency have become important MIPS is usually a good indicator not simplytied to a CPU core tact however it is tied to a specific task such as text search or codecompilation Figure 32 gives a timeline of how CPU frequencies cores on a CPU CPUson a node and nodes in a networked environment have scaled up in about half a century
31 Clustering and Grids Performance Parameters and Basic Models 49
Table 31 Performance of certain selected CPU models
Year CPU model Performance MIPS Tact frequency GHz
2006 AMD Athlon FX60 18938 26
2007 Intel Xeon Harpertown 9368 30
2011 ARM Cortex-A15 35000 25
2011 AMD FX-8150 108890 36
2011 Intel Core i7 2600K 128300 34
2015 AMD A12 Pro-8800B gt150000 34
Fig 32 Timeline of performance indicators in computing hardware
The principles of how CPUs are constructed and how they work have mostly remained thesame [13] but the capabilities have expanded tremendously
The context for tact frequency MIPS and FLOPS is depicted in Fig 33 The followingperformance formula can be used
P D f n1 I n2 (31)
Where P ndash performance in GFLOPS f ndash CPU tact frequency in GHz n1 ndash number ofcores within a CPU I ndash CPU instructions per tact n2 ndash number of CPUs per computingnode Let us consider the integral performance criterion FLOPS in two examples whichinvolve recent server configurations It makes the complex dependency of performancefrom multiple factors evident as the system with the faster CPU is much slower overalldue to less cores and less powerful instruction execution within the cores
Example 32 Let us consider a 2-socket-server with CPU Intel X5675 (306 GHz 6 cores4 instructionstact) P D 306 6 4 2 D 14688 GFLOPS
50 3 Evolution of Clustering and Parallel Computing
Fig 33 Performance parameters of computers
Example 33 We have a 2-socket-server with CPU Intel E5-2670 (26 GHz 8 cores8 instructionstact) P D 26 8 8 2 D 3328G FLOPS
For the performance parameter FLOPS the following nomenclature (K M G T P EZ Y) of the unit prefixes is used
bull KFLOPS KiloFLOPS = 103 FLOPSbull MFLOPS MegaFLOPS = 106 FLOPSbull GFLOPS GigaFLOPS = 109 FLOPSbull TFLOPS TeraFLOPS = 1012 FLOPSbull PFLOPS PetaFLOPS = 1015 FLOPSbull EFLOPS ExaFLOPS = 1018 FLOPSbull ZFLOPS ZettaFLOPS = 1021 FLOPSbull YFLOPS YottaFLOPS = 1024 FLOPS
To put these numbers into perspective The AMD Carrizo-based FX-8800P notebook CPUfrom 2015 which contains four cores and an R7 GPU which operates at a tact of up to34 GHz reaches around 839 GFLOPS An AMD Radeon R300-based R9 Fury GPU from2015 achieves about 7ndash9 TFLOPS with vectoring of operations ie the application of anoperator over multiple elements in a vector Anything in the higher TFLOPS range andabove requires parallel multi-processing or clustering architectures
31 Clustering and Grids Performance Parameters and Basic Models 51
Speedup and effectiveness of computing processes Factors of speedup and effective-ness in grids are computed as follows
An DT1
Tn En D 100
An
n(32)
Where T1 ndash computing time for a math-log problem with use of only one CPU Tn ndashcomputing time of the solution parallelised on n processors or threads An ndash speedup factorEn ndash effectiveness for speedup on n CPUs in
An example for a section distribution by task parallelisation and the influence ofcluster communication exchanges by message passing between the processors or threadsis depicted in Fig 34 The computation time gain is possible only due to higher p=s ndashratio within a parallelised task (a math-log problem) The time estimations are as followsrefer to Eq 33
T D s fnot showng
T D s C p fag
T D s Cp
nfbg
T D s Cp
nC k n fcg
e D 1 p
(33)
s1
s1
s1 K K
s2
s2
a) Sequential workflow
b) Paralleled workflow
c) Paralleled workflow with threads andnetwork exchanges considering
p1 p2 p3
p3
p3
p2
p1
p2
p1
s2
Fig 34 Sections distribution by a math-log problem parallelisation and the influence of clustercommunication (exchanges) by message passing
52 3 Evolution of Clustering and Parallel Computing
Where T ndash overall computing time s ndash sequential part of a task (percentage)p ndash potentially parallelised part of a task (a math-log problem) ie on n threads or CPUse ndash part for sequential computing time k ndash negative influence of communication bymessage passing between CPUthreads (this component can also be neglected k D 0)
Amdahlrsquos Law One of most appropriate and useful approximations for the speedupfactor is the one defined by G M Amdahl in 1967 [9]
T D 1
1 1 p C p
An D1
1 p C pn
1
1 p
Amax D1
1 p
Ank D1
1 p C pn C k n
(34)
Where p ndash potentially paralleled part of a math-log problem n ndash number of availableCPUsthreads k ndash negative influence of communication by message passing betweenCPUsthreads (this component can also be neglected k D 0)
Example 34 Let us consider a math-log problem with an overall compute time ofToverall D 20 h a serial critical compute time of Tser D 1 h (ie 5 ) and a parallelisedcompute time of Tpar D 19 h (ie 95 ) Furthermore let the maximum speedup factor beSpeedupMAX D 20 This is a typical scenario for a scientific computing problem Thenby n D 10 processors (threads) one can derive p D 095 Speedup D 1=1 095
C 095=10 D 1=005 C 0095 D 69 lt SpeedupMAX The results means that outof a theoretic maximum of ten-fold parallel execution only six-dot-nine-fold can beachieved On the other hand with n D 95 processors (threads) the speedup grows toSpeedup D 167 only meaning a reduced effectiveness of only one quarter
One can obtain the following graduated depiction of the speedup factor (Fig 35) Thereare some criticism points regarding this realistic model too pessimistic representationof the parallel computing status But other models talk a lot also about the saturationeffects especially due to communication processes within a cluster between the processors(threads) and energy losses (in form of redundant warm waste heat)
Barsis-Gustafson-Law This law of E H Barsis and J Gustafson proposed in 1988 isfrequently used as alternative compared to Amdahlrsquos law Consider the following Eq 35
1 D 1 p C p (35)
31 Clustering and Grids Performance Parameters and Basic Models 53
25
25
20
15
10
5
0
Threads n
Speedup A(np)
0 100 200 300 400 500
p=05
p=05 p=075 p=09 p=095 p=08
En100
2
15
1
05
00 100 200 300
Threads n
400 500
Fig 35 (a) Speedup vs effectiveness (b) Amdahlrsquos speedup by different p-values PessimisticAmdahlrsquos model for the speedup factor depending on p D 0 5 0 95 saturation effect no moreprofit due to increasing of n ndash number of threads
It decomposes an execution time T into a part which can be parallelised Tp as knowntime for parallel computing and a part which cannot for instance startup or memoryallocation Ts as known time for sequential computing Then the speedup factor iscomputed as shown in Eq 36
Ts D 1 pTp C pTpn
An p D Ts=Tp D 1 p C pn D 1 C pn 1(36)
Example 35 The following example shows how to calculate A according to the paralleli-sation method described by the Barsis-Gustafson law
p = 80 n = 11 CPUs A11 = 1 C 08 (11 1) = 9
n = 31 CPUs A31 = 1 C 08 (31 1) = 25
n = 71 CPUs A71 = 1 C 08 (71 1) = 57
n = 101 CPUs A101 = 1 C 08 (101 1) = 81
Therefore we conclude Amdahlrsquos Law is too pessimisticA typical cluster from Technical University of Chemnitz with 530 nodes called CHiC
is depicted in Fig 36 CHiC nodes run Linux are connected with Infiniband and due tonot having any disks share a Lustre filesystem which spans 160 disks On this kind of
54 3 Evolution of Clustering and Parallel Computing
Fig 36 Fibre glass techniques for CPU coupling (FC ndash Fibre Channel) FC ports offer approximatedata rate = 4 bis 16 GBits performance max 100 GFLOPS per CPU CHiC ndash a powerful cluster[21]
supercomputer consisting only of networked standard computers applications are placedand scheduled according to the beforementioned laws of parallel computing [21]
Karp-Flatt Metric The Karp-Flatt metric (e) is a measure of parallelisation of code inp parallel processors and was proposed in 1990 by A H Karp and H P Flatt [11] Thismetric exists in addition to Amdahlrsquos Law and the Barsis-Gustafson law as an indicationof the extent to which a particular source code for one CPU is parallelised The valueof e (the unknown partpercentage for sequential computing time) can be approximatedon the basis of the metric via known speedup values for different CPU number p andtimes estimations Tp Seven main characteristics need to be distinguished as input for thecalculation
bull A ndash measured speedupbull N gt 1 ndash number of CPUbull T1 ndash time for particular source code for one CPUbull Ts ndash sequential computing timebull Tp ndash parallelised part timebull e ndash part for sequential computingbull p ndash parallelised computing part
In order to estimate the speedup factor Eq 37 needs to be solved
31 Clustering and Grids Performance Parameters and Basic Models 55
T1 D Ts C Tp e DTS
T1
T1 D eT1 C 1 eT1I
TN D Ts C1
NTpI
TN D eT1 C1
NT1 eT1I
A DT1
TN Y D
1
AD
TN
T1I
1
AD Y D e C
1
N1 e
A D Œe C1
N1 e1
(37)
Then we consider responding to value e by solving Eq 38
1
AD e1
1
N C
1
NI
e1 1
N D
1
A
1
NI
e DŒ 1
A 1N
Œ1 1N
D 1 p
(38)
Example 36 We would like to define herewith the value e (refer to formula 39) ie thenormally unknown part for sequential computing time for a math-log problem on the basisof the Karp-Flatt metric Referring to Table 32 (pos 9) the following three parallelisationgrades are given
Number of CPUs n D 100 measured speedup A D 10 1=A D 01 e D 01
001=1 001 D 009=099 D 00909 e D 91 it can be for parallelised p D 91 Number of CPUs n D 100 measured speedup A D 25 1=A D 004 e D
004001=1 001 D 003=099 D 00303 e D 303 it can be parallelised forp D 97
Number of CPUs n D 100 speedup A D 66 1=A D 0 0151 e D 00151 001=
1 001 D 00051=099 D 00052 e D 052 it can be parallelised for p D 995 Considering the previous formulae and Table 32 we can obtain the next useful formula
(39) for the p criterion
56 3 Evolution of Clustering and Parallel Computing
An gt 1
eAn n D 1 p
D
1An
1n
1 1n
p D1 1
An
1 1n
DAn 1
An Ann
DAn 1
An En100
(39)
Example 37 Let us consider the following example The number of CPUs should ben D 100 the speedup A D 66 and the effectiveness En D 66 Then the math-logproblem can be parallelised for the p ratio p D 661=66066 D 65=6534 D 0995
(compare to Example 35)
Moorersquos Law The authorship of the law belongs to Gordon Moore (born 1929)co-founder of Intel Moorersquos Law is known since 1965 and for more than 50 years ithas been holding with no faults It means the exponential growth of the following valueswhich characterise electronics and IT branches
bull CPU chip complexity N (up to 109 transistors)bull Computer tact frequency f (up to 35 GHz)bull Computer performance P (nowadays typically gt100GFLOPs)
Moorersquos Law regarding to the chip complexity is depicted in Fig 37 The values on theY-axis are given in logarithmic scale The next integration degree will reach 10 billiontransistors
But there are some further phenomena which are not commonly associated with thislaw Moorersquos Law is also true for the extrapolation in the backwards direction into theearly days of computing In fact Moorersquos Law extrapolation can be extended down tothe year 1900 towards the former element basis in electronics electro-mechanical relayselectronic tubes transistors IC VLSI as it was depicted in Fig 38
Speedup model overview Table 32 illustrates the set of integrated models and approx-imations of speedup factors which are typically used for distributed (parallel) computingThe table includes the already presented models together with additional ones Theapproximations of the An speedup factor are given with a dependency on the criterian p k These are the mostly used models and laws including Amdahlrsquos (1967) Groschrsquos
31 Clustering and Grids Performance Parameters and Basic Models 57
Fig 37 Moorersquos Law chip complexity (Source it-materialde)
Fig 38 Moorersquos Law extrapolation backwards
58 3 Evolution of Clustering and Parallel Computing
Table 32 Overview on speedup models
SpeedupfactorAn D T1
TnSpeedup model Conventions Title of an empirical model
1 An Dp
n The type of math-log problemis not considered
Groschrsquos law (1965)
2 An D nb The type of math-log problemis not considered
Generalised Groschrsquos law(05 b 1)
3 An D n The type of math-log problemis not considered
Proportional Amdahl law forp D 1 s D 0
4 An D log2n The type of math-log problemis not considered
Logarithmic Law
5 An D 11pC
pn
05 p 0999 Amdahlrsquos Law (1967)
6 An D1
1pCpn Ckn
05 p 0999 k 104 105
Corrected AmdahlrsquosModel with inter-processorcommunication considering
7 An D 2n D 70 =r
The type of math-log problemis not considered r D 1 2 characterises inter-processorcommunication losses
Empirical law ldquo69 - 70 ndash 72rdquofor CPU-number n whichprovides double speedup ofcomputing time
8 An D
1 p C pn05 p 0999 k D 0 Barsis-Gustafson-Law (1988)
9 An gt 1eAn n D 1p
e D 1 p ndash the unknown partfor sequential computing time05 p 0999 k D 0
Karp-Flatt-Metric (1990) forAmdahlrsquos orBarsis-Gustafson-Law
Barsis-Gustafsonrsquos (1988) Moorersquos law (1965 or exponential model) and some furthersuitable models such as the 70 -law [9 11] The evaluation of the coefficient p in theequations can be realised via the Karp-Flatt metric (1990)
A generalised graphical comparison of speedup factors is depicted in Fig 39 Themost-used models are shown a trivial one (3) as well as an optimistic one by Barsis-Gustafson (8) ie more realistic and Amdahl (5) ie a pessimistic one refer to Table 32(3) (5) (8)
Simulation Scenario For the hardware basis (Fig 310a) offered at Dresden Universityof Technology [15] the following own results (Table 33) on speedup have been obtainedIt was a voluminous experiment in November 2006 aimed at the simulation of signalpower propagation of WLANWiMAX networks through complex 2D environmentswhich appeared as maps of the obstacles with given material features
The simulation has been realised with use of CANDY software and web servicesfor SSL access to MARS The following results have been obtained (Fig 311 refer toTable 33) These results can be approximated with formula (310) compare Groschrsquos law
31 Clustering and Grids Performance Parameters and Basic Models 59
Fig 39 Speedup models ndashdifference between optimistic(3) and pessimistic view (5)
Fig 310 (a) Hardware basis High-performance computing cluster MARS SGI Altix 4700 TUDwith 1024 cores possesses the performance 131 TFLOPS (b) Up-to-date hardware basis TAURUSBull HPC-Cluster with 137 TFLOPS Hardware basis High Performance Computing at TUD [15]
An DT1
TnD n˛ T1 D 8021s ˛ 095 (310)
Example 38 The new hardware basis in the same institution is called TAURUS Bull HPCcluster This cluster is more powerful than the formerly leading MARS placed at globalrank 66 at its inauguration and has nowadays the following features (Fig 310b)
bull Island 1 4320 cores Intel E5-2690 (Sandy Bridge) 290 GHzbull Island 2 704 cores Intel E5-2450 (Sandy Bridge) 210 GHz as well as 88 NVidia Tesla
K20x GPUs
60 3 Evolution of Clustering and Parallel Computing
Fig 311 Computing time and speedup factor in depending on threads number obtained on themulti-core high-performance computer MARS TU Dresden (Basis ndash CANDY Framework 2006)
Table 33 Computing time fora complex simulation task ofWLANWiMAX propagation
Number of threads Computing time s Speedup factor An D T1Tn
1 8021 10
2 4163 19
5 1749 46
10 908 88
20 471 170
30 321 250
55 181 443
70 144 557
bull Island 3 2160 cores Intel X5660 (Westmere) 280 GHzbull Symmetric Multi-Processing (SMP) nodes with 1 TB RAMbull 1 PB SAN disk storagebull Bullx Linux 63 based on Red Hat Enterprise Linux batch system Slurmbull 137 TFLOPS total peak performance (without GPUs)
Example 39 The most performant cluster of the world is depicted in Fig 312 TheTianhe-2 or ldquoHeaven Riverrdquo (Milky Way) originates from Guangzhou in the PeoplersquosRepublic of China The common costs for the cluster can be evaluated to be approximately24 109 Yuan (equal to USD 390 106) The peak performance is P D 33PFLOPSThe square size S D 720 m2 belongs to the cluster Surely the power consumption iscorrespondingly very high about 17 24 MW But also a very high PUE value is to benoted The nodes of the cluster use a specific operating system Kylin Linux which has alsoinfluenced Ubuntu Kylin to become recommended as reference system for many Chinesedeployments until 2018 The available compilers are as follows Fortran C C++ JavaOpenMP MPI 30 Tianhe-2 possesses the following architecture
bull 32000 CPUsbull 48000 GPUs as programmable co-processors
31 Clustering and Grids Performance Parameters and Basic Models 61
Fig 312 The most powerful compute cluster world-wide Tianhe-2 (Sources top500orghpcwirecom photo onlinezeitung24de)
Table 34 Computing system performance comparison (Status November 2015)
Cluster or gridMaximum performancePFLOPS
Multiplicity (given inldquoMARS unitsrdquo)
Tianhe-2 (a supercomputer from GuangzhouChina)
3386 2605
Titan (Tennessee USA supercomputer upgradefrom Jaguar)
1759 1353
BOINC (grid hosted at Berkeley University ofCalifornia USA)
9 692
Juqueen (FZ JuumllichIBM) 50 384
SuperMuc (Leibniz data centre in Munich) 28 215
TAURUS (hosted at TU Dresden) 103 79
MARS (TU Dresden 2006) 0013 1
bull 1375 TiB of RAM of which 1000 TiB is accessible by the CPU and 375 TiB by theco-processors
bull 124 PB hard disk capacity
The total number of cores exceeds three million and achieves a combined performanceof 3386 GFLOPS The predecessor in the global ranking top spot has been the Titansupercomputer in the USA with ldquojustrdquo 1759 GFLOPS
SMP architectures with large RAM capacities gains in its deployment nowadays moresympathisers than the NUMA (Non-Uniform Memory Access) with the offered uniqueaddress spaces as well as correspondingly the cache-coherent NUMAs A performancecomparison is given in Table 34 Herewith some worldwide known clusters from the
62 3 Evolution of Clustering and Parallel Computing
global top-500 list (TOP500) as well as grids are referred in correspondence to the abovementioned performance of MARS and TAURUS systems The MARS performance isgiven as canonical base unit Most of the clusters about 98 run Linux whereas gridsallow for heterogeneous operating systems in particular desktop grids such as BOINCThe performance values are measured with the LINPACK benchmark a Fortran librarywith routines to solve linear algebra equations
32 Performance-Energy-Price Trade-Offs in Clusters and Grids
Trend to low-cost and low-energy computing nodes A new trend to low-cost and low-energy computing nodes based on cheap devices in particular cheap and fanless on-boardmicroprocessors (RISCARM) should be considered nowadays as a serious alternativeto expensive computing devices within Internet of Things (IoT) a term describing avision of ubiquitous access among connected devices On top of the IoT an Internetof Services (IoS) with digital and physical services can be constructed The IoS is arelated vision which for most applications hides the hardware The deployment of low-cost and low-energy computing nodes such as those with Arduino Raspberry Pi or IntelEdison processors leads to a significant increase of energy-efficiency outcomes as well as atechnologically important new step towards a realisation of the IoT Often these connecteddevices are seen as Fog Computing backbone to an even larger IoT which also involvesstationary and mobile sensors such as mobile phones and heartbeat belts [2 27]
Trade-offs Scenarios for the so-called Fog Computing within the IoT are steadily goingto gain in importance in the mid-term Instead of using applications and services withheavy-weighted processors and VMs agile and energy-efficient on-board microprocessorsshould be operated See the view of future transfer from CloudsIoS to the Fog Comput-ingIoT (Fig 313) Surely the deployment of low-cost and low-energy computing nodesbased on on-board microprocessors can be used to build powerful clusters as well Theselead to an appropriate resource use in the frame of a given math-log problem
On-board microcontrollers But none of the above-mentioned computing systems isenergy-efficient enough The electricity consumption is measured in the MWh areaEnergy-efficient solutions can be provided via small low-cost and low-energy on-boardprocessors The electricity consumption surrounds in this case at most the kWh areaLow-energy home intelligent nodes (3ndash10 W) for private cloud solutions file serversweb servers multimedia home centres and similar use cases can be operated with suchmicrocontrollers as the trade-off solution They offer a cheap alternative and symbolise astep-by-step shift towards the IoT
Example 310 Herewith a small example addressing the discussed trade-offs A ldquosuper-computerrdquo with 64 cheap Raspberry Pirsquos und two Lego racks is depicted in Fig 314 This
32 Performance-Energy-Price Trade-Offs in Clusters and Grids 63
VM VM VM VM
VMM
- Universal Service XaaS
Cloud Computing
On-board μ-Nodes
Fog Computing
Reliable VM orlow-energy μ-Node
Trade-offs
- VM Monitor- Dedicated VM
Reliability and QoSData Security and PrivacyAnonymityEnergy ConsumptionOperating Expenses (OPEX)
Raspberry Pi
μ
μ
μ
μμ
μ
ArduinoIntel Edison
VM VM
Fig 313 Energy-efficient on-board computing nodes as a basis for distributed computing withsufficient performanceenergyprice trade-off
Fig 314 Energy-efficient Raspberry Pi cluster with 64 CPUs (Source pro-linuxde)
low-energy cluster (64 35 W maximum 025 kW) is built by using low-cost and energy-efficient on-board microcontrollers The small but smart Raspberry Pi cluster for parallelcomputing offers the following features
bull DC supply through USB 35 WCPU 700 MHzbull Energy-efficient resource provisioning
64 3 Evolution of Clustering and Parallel Computing
Fig 315 Data centers of Google internal view (Source Google)
bull SD card as external disk drivebull Low-power data transfer and exchange via Ethernet LANbull Raspbian as operating system
Energy-efficient data centers of Google Around 2011 the trend of ldquoGreen ITrdquo wastriggered by increasing energy demand and prices and a general awareness of computingusers The data and computing centers have to be built step-by-step in colder regionsof the earth The data centres of Google achieve the Power Usage Effectiveness (PUE)of 112 due to further optimisation of hardware waste heat recycling systems andbuilding construction features like improved air circulation reuse of waste heat andother techniques [6] This means that only 12 of energy required for computingwas used not by servers but by other services like conditioning energy distributionlighting surveillance systems etc (Fig 315) Hence note that the value of Power UsageEffectiveness (PUE) of 10 is only possible in theoretic ideal cases It means that there arenot any additional energy losses or waste heat what is contradictive indeed to the classicalthermodynamic theory
33 Resource Management in Clusters
First three single-system cluster management systems which integrate with the operatingsystem will be presented Then a resource management placement and schedulingframework which runs on top of an operating system will be compared
MOSIX OpenMosix and OpenSSI cluster management While most clusters includ-ing Beowulfs only share the filesystem among nodes single-system image (SSI) clustersshare the entire operating system instance including processes virtual memory open files
34 Application Management in Clusters 65
sockets and inter-process communication In such systems applications get access to morecompute resources like in SMP or multi-core environments only with added networklatency The broad existence of multi-core processors has caused a decline in managementsystems for SSI clusters but as they can still be useful three such systems shall bepresented here MOSIX OpenMosix and OpenSSI both derive from the Linux operatingsystem kernel The active development phase of OpenSSI was from 2001 to 2010 andof OpenMosix from 2002 to 2008 following as derivative (fork) on MOSIX from 1999which is still actively maintained today in the form of MOSIX2 and MOSIX4 A referencedeployment of MOSIX runs a private production-level cloud consisting of 11 SSI clustersin particular for computer science life sciences and medical school applications Theclusters combine 205 nodes with an average of 35 active nodes and 200 processorcores
Resource management placement and scheduling with Mesos Apache Mesos imple-ments modified versions of typical application computing frameworks such as HadoopSpark Kafka or Elastic Search When the application submits tasks to be processed theyare placed close to the data without the application having to know the data locationFurthermore Mesos is fault-tolerant and safe in the sense that tasks can be executedas isolated processes using the Linux containers interface It uses ZooKeeper to ensureconsensus among all nodes in the cluster and it offers a web interface to check the clusterstatus
34 Application Management in Clusters
Once a non-SSI cluster its nodes and its resources are managed the applications runningon it need to be managed as well As opposed to an SSI cluster a failure of a node impliesthe failure of one instance of the (parallelised) application and appropriate migrationand restart techniques are required to avoid the propagation of the failure to the userIn this section three application managers for cluster environments will be comparedTheir common aim is easy deployment fault-tolerant and resilient execution of parallelisedsoftware applications
Kubernetes Fleet and Pacemaker Kubernetes is a container cluster manager developedby Google which makes the cluster appear as a single system despite not being anSSI cluster It eases the deployment maintenance and scaling of application partswhich are packaged as executable Docker containers Google uses it behind the GoogleCompute Engine (GCE) but it is also used by other hosting providers includingTecTonic
Fleet extends Systemd a daemon which initialises and supervises application pro-cesses towards multiple nodes in a cluster Again the application is supposed to bepackaged as Docker containers Fleet ensures that a minimum number of container
66 3 Evolution of Clustering and Parallel Computing
instances is running across all nodes in the cluster and starts new instances in case ofan application or node failure Fleet uses a configuration daemon called Etcd to ensureconsensus among all nodes and to implement discoverable nodes By placing containerinstances on different nodes and assuming a fault-tolerant load balancer the overallavailability of services offered by the applications is increased
Pacemaker is a cluster manager aiming at high availability of applications Applicationsare replicated onto two or more nodes with activepassive standby functionality oractiveactive failover and a subsequent recovery by application migration Pacemaker isdeveloped by Cluster Labs and used for instance by the German flight safety companyDeutsche Flugsicherung (DFS)
Apart from these complex systems simple tools exist to manage commands on clustersAmong these tools ClusterSSH Ansible and Puppet are popular to replicate installationand configuration instructions to all nodes in the cluster
35 Application Management in Grids
In this section two grid systems will be presented BOINC and OurGrid The criteria whichled to the selection of these two grid systems are recent or ongoing development and publicavailability Thus interested readers are welcome to download the software and connecttheir own computers to an existing grid or even open a new grid for others to join Bothgrids offer computer capacities for various applications
BOINC desktop grid BOINC is a volunteer computing project aimed at contributingcompute resources (ie spare CPU cycles) to scientific projects [28] BOINC is hence alsoa grid platform for scientific projects and HPC developed at the University of Berkeleyfor free distribution licenced under the GPL The availability is offered for the followingoperating systems Windows Linux Mac OS X Android and BSD The BOINC platformprovides an unlimited computing power of up to hundreds of thousands of computersworld-wide coupled via the Internet The cooperation is organised in the form of projectsrunning atop The architecture of BOINC is given in Fig 316 The main components arethe BOINC daemons long-running services which interact with the BOINC clients byexchanging data
Most of the scientific computing grids work to the profit of universities or otherscientific institutions BOINC is a well-known grid around the world due to its combinedstructure client-server (C-S) and peer-to-peer (P2P) The servers distribute the applicationpackages to the clients In general these ldquoclientsrdquo serve the architecture themselves in aP2P topology The client applications calculate intensively (usually 2-40 h per package)and report the solutions to a main structure (the server) Optionally another solution
35 Application Management in Grids 67
Fig 316 BOINC architecture [12] (Sources gclcisudeledu boincberkeleyedu)
for the client receives a verification According to status of 2015 the BOINC gridpossesses [28]
bull Nowadays approximately 250000 persons and 850000 computers (notebooks tabletsand other devices) are involved in a cooperation with BOINC
bull Overall performance of the grid system BOINC 9 PFLOPS (refer Table 34)
Compared to these metrics the performance of some super-computers from the bi-annualglobal top-500 list is as follows
bull Tianhe-2 (ldquoMilky Wayrdquo ldquoSky Riverrdquo China) with 3120000 cores ndash 3386 PFLOPSbull Titan (USA) with 560000 cores ndash 1759 PFLOPSbull Mira (USA) with 786000 cores ndash 858 PFLOPS [25]
Anyone can run the BOINC servers If the server is public the results must be alsopublished to prevent the abuse and misuse An interesting idea is the use of BOINC withincompanies
bull An internal BOINC server distributes in-house applications to the employeesrsquo comput-ers
bull More effectiveness because the desktop systems are usually not enough loadedunchallenged eg usage of Word Outlook CRM in the everyday workflow
68 3 Evolution of Clustering and Parallel Computing
Fig 317 BOINC client-server interaction (Sources gclcisudeledu boincberkeleyedu)
The interaction protocol between a client and server (ia PC notebooks tablets smart-phones and other devices) is depicted in Fig 317 The error-free interaction uses fivephases
Top-10 of the most popular projects In cooperation with BOINC a number of piggy-backed projects have been supported The top ten of the most popular projects are asfollows
1 SETIHome ndash Analysis of a series of radio telescope data from space for thepurpose of searching for extra-terrestrial civilisations (Search for Extra TerrestrialIntelligence)
2 EinsteinHome ndash Tests of the hypothesis of Albert Einstein about gravitation wavesand search for radio- and gamma ray pulsars
3 World Community Grid ndash Assistance in the search for medicaments for seriousdiseases such as cancer HIV AIDS the calculation of the 3D structure of proteinsand a lot of other projects (organiser ndash IBM)
4 RosettaHome ndash Calculation of the 3D folding structures of proteins based on theamino acid sequences for the treatment of cancer HIV AIDS Alzheimerrsquos diseaseanthrax (Siberian ulcer) etc
5 MilkyWayHome ndash development of a precise 3D model of the stellar streams in ourgalaxy (Milky Way)
6 Climate Prediction ndash Research and prediction climate on earth7 PrimeGrid ndash Search for diverse prime values8 SIMAPHome ndash Creating a database of proteins for bioinformatics9 CosmologyHome ndash Search for a model which adequately describes our universe
and is consistent with current data in astronomy and particle physics10 Collatz Conjecture ndash Studies in the math specially to test the hypothesis of Lothar
Collatz also known as ldquoproblem 3n + 1rdquo
35 Application Management in Grids 69
Legend
Project Back-end
BOINC Components Project specific Components
ProjectScience
DatabaseBOINC Back-end Interface
BOINC DaemonScreen-Saver Engine
BOINC Software
BOINC Manager
ParticipantrsquosComputerScreen-Saver
BOINCDatabase
A BOINCPoweredProject
Participant
ProjectDatabase
Science Application
API
BOINC Server Complex
DataServer(s)
SchedulingServer(s)
Web Server
BOINC Web Pages
Project Web Pages
Fig 318 Advanced BOINC-II architecture [16]
In total more than 40 projects can be chosen by volunteering participants to contributespare compute resources to
Example 311 Malaria Control is a popular project which runs on top of BOINC-II thelatest generation of BOINC Its goal is to gather and analyse information about the Malariadisease
The advanced BOINC-II architecture [16] is depicted in Fig 318 A new BOINC APIseparates screensaver into a standalone program The details of the use of the science
70 3 Evolution of Clustering and Parallel Computing
Fig 319 An OurGrid federation with three peers
applications (eg for malariacontrolnet) the BOINC-II specific components as well as ofthe project specific components are discussed in [16]
OurGrid OurGrid developed since 2004 by the Federal University of Campina GrandeBrazil federates networks of connected computers to support the distributed parallelexecution of jobs and tasks in a grid The federation happens with a peer-to-peer topologyusing the Extensible Messaging and Presence Protocol (XMPP) Jobs are executed asJava or system-level virtual machines as sandbox in order to isolate them from each otherand from the software and data on the host computers [5] Each peer in the federation isa network of connected computers consisting of worker and broker nodes The discoverymechanism among all the nodes relies on XMPP as well Jobs are submitted along withscripts executables data and a job description file which outlines the tasks of a job Aunique feature of OurGrid is the implementation of the Network of Favours reputationmechanism to ensure fairness and to avoid freeriders who consume compute resourceswithout contributing them back at some point Figure 319 shows an example of aninstance of OurGrid across three networks of connected computers which may or may notbe clusters
Desktop computers are suitable as workers because the idleness detector prevents aconflict between interactive use and a high load from the submitted jobs Furthermore thesystem has been designed as opportunistic grid so that failures shutdowns and hibernationswill only interrupt the current task execution without affecting the job as the affected taskwill be restarted Hence OurGrid is suitable to be used to offer both opportunistic gridswith many resources and service grids with high quality of service on the same physicalinfrastructure [3] The OurGrid project is now inactive but the software is still functionalfor setting up further instances
36 Distributed Applications 71
36 Distributed Applications
Whereas in grids the infrastructure is distributed but the application itself is merelyconsisting of offloaded job and task units some applications are truly distributed in apeer-to-peer sense or decentralised in a hub-and-spokes model [22] Representatives ofthese two models will be presented in this section
Distributed blockchains hashtrees and cryptocurrencies A blockchain is a poten-tially large file which contains entries (chronologically ordered blocks) whose contentdepends on previous blocks Due to the size it is possible to distribute parts of the fileto different users With cryptographic methods it is possible to ensure consistency and toprevent forgery in older blocks When such a linear structure is not sufficient hashtreespresent similar characteristics but allow for subsuming multiple blocks under one blockand eventually a whole tree of blocks under one common root There are many interestingapplications resulting from such a globally shared data structure For instance securedblockchains are used to record virtual currency transactions leading to cryptocurrencieswith properties like anonymity and traceability of transactions To regulate the valuedistribution in such a currency the blockchain can only be extended after a compute-intensive effort with a certain difficulty The Eq 311 refers to the profitability to advancea distributed blockchain with a given difficulty referred to in Eq 312
profit D revenue costelectricity C costdifficulty (311)
costdifficulty Dmaximum difficulty
current difficulty
232
hashrate(312)
Example 312 Bitcoin is a popular example of a cryptocurrency which is mined from adistributed blockchain Similar to distributed desktop grids the participants donate CPUcycles for a cause In contrast to the grids however the cause is not directly involvinga global problem solving effort or a citizen science effort but rather the race for thequickest solution of an algorithmic problem which lets the blockchain advance At thesame time a fictive virtual currency coin is yielded The value of such a coin depends alot on perception trust and market dynamics In Bitcoin there has been a steady growthat first followed by an unpredictable development At the same time the production costfor mining has increased a lot due to the nature of the blockchain which requires morehardware resources for each subsequent solution Hence already from an energetic pointof view the effort required to advance is not compensated anymore by a potential gainfrom the virtual cryptocurrency coins Figure 320 outlines the profitability graph overtime It shows that the price (green) surged in November 2013 followed by its declineAt the same time the difficulty to mine (red) increased by several orders of magnitude
72 3 Evolution of Clustering and Parallel Computing
Jul1
1
Pricedifficulty1000000000Difficulty
Price
Oct
11
0k5k10k
15k
20k
25k
30k
Jan
12A
pr1
2Ju
l12
Oct
12
Jan
13A
pr1
3Ju
l13
Oct
13
Jan
14A
pr1
4Ju
l14
Oct
14
Jan
15A
pr1
5Ju
l15
Oct
15
0 U
SD
0G
200
US
D 1
G
400
US
D 2
0G
600
US
D 3
0G
800
US
D 4
0G
1000
US
D 5
0G
1200
US
D 6
0G
Fig
32
0D
evel
opm
ento
fB
itcoi
npr
ofita
bilit
yov
ertim
e(S
ourc
eco
inpl
orer
com
)
36 Distributed Applications 73
Hence the profitability as quotient of the two converged quickly against zero and whenaccounting for the energy cost is already negative
Example 313 Git is an example of a distributed version control system built atop ahashtree Each Git repository contains a directory structure with files File changes canbe performed independently from each other Once changes are committed they and theirassociated metadata records are cryptographically secured against forgery and tamperingThe Git model leads to high scalability in large collaborative file editing efforts includinglarge software development teams
Decentralised and federated social networks Social networks are one of the mainapplications on the Web and on the Internet today They incorporate communicationpatterns between their participants and add useful or convenient functionality such asvisibility management for events a timeline of events as well as add-on applications Theirappearence is either web-based or through communication protocols LinkedIn Facebookand Twitter are examples of the former category whereas ICQ and similar chat systemsare examples of the former one Their commonality is a centralised hosting so that eachmessage is relayed through a potentially distributed physical set of servers but within onelogical organisation In contrast federated social networks allow any participant to choosebetween joining an existing server or running their own server An examples is Diaspora
Example 314 Diaspora is a web-based federated social network which can be run incentralised decentralised and distributed configurations Users sign up at a server calleda pod and receive an account in the form of loginpod They can add contacts (friends)from the same or from other pods Message posts from all contacts are then aggregatedand shown in the timeline of each respective user A typical aggregated Diaspora timelineis shown in Fig 321 The aggregation function fetches the posts from all connected podsorders them chronologically and caches them to increase the scalability and to decreasethe latency for subsequent timeline retrievals
Collaborative real-time applications Whereas web-based social networks cryptocur-rencies and version control systems work inherently asynchronously so that each user candecide when to update the local state from the (potentially increasingly diverging) globalstate there is also a class of distributed applications which works synchronously in real-time Among the most prominent are scalable chat audio and video conferences
An example for a real-time chat application with extensions for audio and videoconversation is the XMPP A second example if WebRTC a web browser overlay overthe conventional Real-Time Communication (RTC) protocol
Example 315 Users of XMPP servers receive fully-qualified accounts with a login nameand a server name in the form of loginserver This way similar to e-mail the serverscan federate so that users from different servers can communicate with each other
74 3 Evolution of Clustering and Parallel Computing
Fig 321 Diaspora timeline with aggregated friend feeds
XMPP defines a core messaging protocol and several extensions for registration binaryattachment transmission VoIP communication and other features The chat protocol isalso known as Jabber and the VoIP protocol as Jingle
Due to the nature of being a communication protocol humans and software applicationscan equally participate in XMPP networks Software components are registered as clientsBy registering their functionality at a discovery service they can also offer servicefunctionality according to the message-oriented architecture paradigm
Example 316 WebRTC negotiates a connection between two users of web browserswith XMPP Jingle as well as the JavaScript Session Establishment Protocol (JSEP) Nocentral server is required for both the negotiation and the subsequent bidirectional datatransmission instead the communication host needs to transmit the dynamically createdendpoint (a URL) to the other participants
37 Conclusions
The scale-up from individual computers to clusters and grids in the past decades thwartsthe ongoing trend towards miniaturisation of computing hardware Nowadays a quad-core mobile phone has a lot more computing power than the original Beowulf cluster with16 nodes and consumes only a fraction of the electric power Still the need for vertical
References 75
performance scale-up remains and through parallelisation becomes a horizontal scale-outoperation into multiple nodes of a system-on-a-board cluster or multiple compute servicesin a grid or cloud With the broad availability of open source software to run private clustersand grids which can be federated with existing public ones supercomputing as well ascomfort computing is now available to every user
References
1 Joel C Adams Jacob Caswell Suzanne J Matthews Charles Peck Elizabeth Shoop and DavidToth Budget Beowulfs A Showcase of Inexpensive Clusters for Teaching PDC In Proceedingsof the 46th ACM Technical Symposium on Computer Science Education (SIGCSE) p 344ndash345Kansas City Missouri USA March 2015
2 F Bonomi R Milito J Zhu and S Addepalli Fog Computing and Its Role in the Internet ofThings CISCO whitepaper 2007
3 Francisco Brasileiro Alexandre Duarte Diego Carvalho Roberto Barbera and Diego Scar-daci An Approach for the Co-existence of Service and Opportunistic Grids The EELA-2Case In Latin-American Grid Workshop Campo Grande Mato Grosso do Sul BrazilOctoberNovember 2008
4 Mario Cannataro Clusters and Grids for Distributed and Parallel Knowledge Discovery InHigh Performance Computing and Networking 8th International Conference (HPCN) Europevolume 1823 of Lecture Notes in Computer Science p 708ndash716 Amsterdam The NetherlandsMay 2000
5 Walfredo Cirne Francisco Brasileiro Nazareno Andrade Lauro Costa Alisson AndradeReynaldo Novaes and Miranda Mowbray Labs of the World Unite Journal of GridComputing 4(3)225ndash246 2006
6 Jeff Dean Designs Lessons and Advice from Building Large Distributed Systems In 3rd ACMSIGOPS International Workshop on Large Scale Distributed Systems and Middleware (LADIS)Big Sky Montana USA October 2009
7 Javier Fabra Sergio Hernaacutendez Joaquiacuten Ezpeleta and Pedro Aacutelvarez Solving the Interoper-ability Problem by Means of a Bus An Experience on the Integration of Grid Cluster and CloudInfrastructures Journal of Grid Computing 12(1)41ndash65 March 2014
8 Bjoumlrn Gmeiner Harald Koumlstler Markus Stuumlrmer and Ulrich Ruumlde Parallel multigrid onhierarchical hybrid grids a performance study on current high performance computing clustersConcurrency and Computation Practice and Experience 26(1)217ndash240 January 2014
9 John L Gustafson Reevaluating Amdahlrsquos Law Communications of the ACM 31(5)532ndash5331988
10 Violeta Holmes and Ibad Kureshi Developing High Performance Computing Resources forTeaching Cluster and Grid Computing Courses In International Conference On ComputationalScience ICCS ndash Computational Science at the Gates of Nature volume 51 of Procedia ComputerScience p 1714ndash1723 Reykjavik Iceland June 2015
11 A H Karp and H P Flatt Measuring Parallel Processor Performance Communications of theACM 33(5)539ndash543 1990
12 Andrew Leaver-Fay Michael Tyka Steven M Lewis Oliver F Lange James ThompsonRon Jacak Kristian Kaufman P Douglas Renfrew Colin A Smith Will Sheffler Ian WDavis Seth Cooper Adrien Treuille Daniel J Mandell Florian Richter Yih-En Andrew BanSarel J Fleishman Jacob E Corn David E Kim Sergey Lyskov Monica Berrondo StuartMentzer Zoran Popovic James J Havranek John Karanicolas Rhiju Das Jens Meiler Tanja
76 3 Evolution of Clustering and Parallel Computing
Kortemme Jeffrey J Gray Brian Kuhlman David Baker and Philip Bradley ROSETTA3 anobject-oriented software suite for the simulation and design of macromolecules Methods inenzymology 487545ndash574 2011
13 Linkfeed Vom Sand zum Prozessor online in German httpgumzodepost171 201514 Seyedeh Leili Mirtaheri Ehsan Mousavi Khaneghah Lucio Grandinetti and Mohsen Sharifi
A mathematical model for empowerment of Beowulf clusters for exascale computing InInternational Conference on High Performance Computing amp Simulation (HPCS) p 682ndash687Helsinki Finland July 2013
15 Wolfgang Nagel and Ulf Markwardt High Performance Computing (HPC) at ZIH HPC Sys-tems Technische Universitaumlt Dresden online httptu-dresdendedie_tu_dresdenzentrale_einrichtungenzihhpchochleistungsrechner 2015
16 Christian Ulrik Soslashttrup Nicolas Maire BOINC II Niels Bohr Institute (CopenhagenDenmark)Swiss Tropical and Public Health Institute (Basel Switzerland) 2014 42p
17 Jong Hyuk Park Laurence T Yang and Jinjun Chen Research trends in cloud cluster and gridcomputing Cluster Computing 16(3)335ndash337 2013
18 A I Petrenko The application of grid technologies in science and education NTUU ldquoKPIrdquoKyiv 2008 143 p in Ukrainian
19 A I Petrenko B V Bulakh and V S Khondar Semantic grid technologies for science andeducation NTUU ldquoKPIrdquo Kyiv 2010 178 p in Ukrainian
20 A I Petrenko S Ya Svistunov and G D Kiselev Grid Technologies Practical Course NTUUldquoKPIrdquo Kyiv 2011 448 p in Ukrainian
21 Wolfgang Rehm and Arnd Meyer TU Chemnitz HPC Cluster CLiCCHiC online httpswwwtu-chemnitzdechic 2015
22 Alexander Schill and Thomas Springer Verteilte Systeme ndash Grundlagen und BasistechnologienSpringer-Verlag second edition 2012 433 p in German
23 Volkmar Sieh Performance metrics online httpwww3informatikuni-erlangendeLehreCPUSS2012multiprocessorpdf 2012
24 Larry Smarr and Charles E Catlett Metacomputing Communications of the ACM 35(6)44ndash52June 1992
25 Erich Strohmaier Jack Dongarra Horst Simon and Martin Meuer The 45th TOP500 Listonline httpwwwtop500orglists June 2015
26 Andrew S Tanenbaum and David J Wetherall Computernetzwerke Pearson Studium fifthedition 2012 1040 p in German
27 R van Kranenburg The Internet of Things A critique of ambient technology and the all-seeingnetwork of RFID Pijnacker Telstar Media 2008 62 p
28 Aacutedaacutem Visegraacutedi Joacutezsef Kovaacutecs and Peter Kacsuk Efficient extension of gLite VOs with BOINCbased desktop grids 2014
4Cloud Computing Virtualisation Storageand Networking
Keywords
Service models bull Internet of Services (IoS) bull Software-as-a-Service (SaaS) bullInfrastructure-as-a-Service (IaaS) bull Platform-as-a-Service (PaaS) bull Virtualisa-tion bull Software-Defined Networking (SDN) bull Security and availability bull Cloudbackup and backup clouds bull Redundant Array of Independent Clouds (RAIC) ndashstripes and parity based dispersion bull Virtual Telecommunication Engineering Offices(VTEO) bull Mobile cloud access bull Network and online storage integration
In recent years networking technologies obtained large success regarding to data rate(WDM MPLS 10GbE) mobility (HSDPA LTE in mid-term 5G) universality andaccessibility of computing services [8] The pervasiveness of services helped to make theIoS become reality and practically accessible for multiple users and appliances Amongthe most prominent service classes in the IoS are Cloud Computing services which aredelivered to their users on demand through desktop mobile and web applications as wellas other forms of user interfaces Modern Internet connections with high bandwidth andlow latency allow a global-scale delivery and complement with attractive (mobile) servicesin the same way and Quality of Service (QoS) the services which have been mostlythe domain of local networks such as corporate e-mail or scientific compute grids Thediscussed information technology paradigm for serving resources and applications to thinclients represented frequently via only low-performance appliances and devices is calledcloud computing [818] As one of the most important IoS forms we will discuss below thebasic cloud computing technologies in the first section The subsequent sections will thenpresent details about virtualised compute networking and storage services which togetherform the core set of resource services available through cloud infrastructure services
copy Springer Fachmedien Wiesbaden GmbH 2017A Luntovskyy J Spillner Architectural Transformations in Network Services andDistributed Systems DOI 101007978-3-658-14842-3_4
77
78 4 Cloud Computing Virtualisation Storage and Networking
41 Clouds Technology Stack Basic Models and Services
Floating in the clouds From a service consumer perspective cloud computing offersmany advantages Many of the offered products and services cater to the traditional desireof users to get anything (information resource and application services as well as products)with a snip of their fingers Many users would like to float in the clouds figurativelymany of them with a cellular smartphone and get anything on demand without delay andwithout cumbersome registration and payment processes (Fig 41)
There are many statistics about how prevalent cloud services a subset of these on-demand services are Certainly a large majority of users is unaware about whethera functionality is completely contained within a device or either aided or completelyprovided by external services Estimations exist about the habits of users
bull 99 of all emailsbull 25 of all notesbull 33 of appointmentsbull all images in social networksbull all online storages
Fig 41 Anything serviced on demand from the clouds
41 Clouds Technology Stack Basic Models and Services 79
Fig 42 Cloud architecture (own representation HPNW denotes High-Performance Network)
These ratios are driven by online services in particular SaaS but also ndash in particular forstorage ndash IaaS
A general architecture and overview for cloud services is given via Fig 42 This holisticarchitecture extends beyond the scope of a single service provider but also omits detailssuch as multi-site replication of services
Cloud computing can thus be defined to be the on-demand and pay-per-use applicationof virtualised IT services over the Internet or within the IoS The key features of cloudcomputing based on the National Institute of Standards and Technology USA (NIST)definitions [8 24] are as follows
bull on-demand self-service with instant delivery on requestbull broadband network access (multimodal all-in-IP)bull resource pooling and rapid elasticitybull measured and optimised service for reliable QoS guaranteesbull service-oriented Internet (Service-Oriented Architecture (SOA) IoS)bull Everything-as-a-Service (XaaS) also represented in Fig 43
80 4 Cloud Computing Virtualisation Storage and Networking
Fig 43 Cloud computing as a pyramid model based on NIST
Fig 44 Difference between hybrid public and private clouds
Public cloud computing platforms are run by commercial providers and by researchorganisations and to a lesser degree by individuals for instance volunteers in self-organised communities Furthermore private and hybrid cloud environments are runwithin company and institution departments Prominent examples of commercial cloudcomputing platforms are as follows Amazon Web Services Oracle Cloud WindowsAzure IBM Softlayer and BlueMix Google Cloud Platform The common organisationtypes of clouds are given below (Fig 44)
The difference between public hybrid and private clouds is presented in greater detailin Table 41
Clouds as new information technology foundation In cloud environments access tocomputing resources (compute storage and network) is performed with the aid of basic
41 Clouds Technology Stack Basic Models and Services 81
Table 41 Hybrid public and private clouds
Private cloud Public cloud
Customer-specific operated by the cus-tomer cloud environment
Owned by an IT service located and oper-ated by this cloud environment
Access limited (customer himself autho-rised business partners)
Access via Internet
Access via Intranet Flexible and easy use by subscription
Hybrid Cloud
Combined Private and Public Cloud
web services most often based on the Hyper-Text Transport Protocol (HTTP) [19]Three general service classes are typically subsumed when talking about cloud computingSaaS is the simplest model with interfaces supporting service-oriented applications whichprovide access to functionality and data delivered through the cloud as frontend PaaSis used for offering to developers an integrated environment for development andortesting of applications as testbed The model IaaS is applied for offering virtualisedresource services in remote computing and networking structures inter alia due to use ofthe remote servers Storage-Area Network (SAN)Network-Attached Storage (NAS)virtual machines and switching equipment The set of functions available through thesecloud services are provided for thin client access to the virtualised resources and multi-tenant hosted applications with non-transparent internal structure The aims are diverseand include high performance of certain routines resources and time-consuming tasks aconsolidation andor partitioning of available physical resources as well as integration ofdesktop mobile and web applications for enterprise informational systems in scenarios ofEnterprise Application Integration (EAI) [36] Load balancing and function distributionbetween cloud computing and conventional IT management are depicted in Table 42 Theproviders of these services within the wider IoS offer to their end-users multiple attractiveservices on different hierarchical levels The table depicts the representation which hasbeen established in accepted best practices documents of important industry players
The purpose of the creation and maintenance of different service-oriented applicationsis to deliver easy-to-use standardised Application Programming Interface (API) end-points for multiple target platforms Frequently the internal structure of a cloud staysnon-transparent for the end-users [28] The users are forced to outcrop from the full-trustposition to their own cloud provider or even to multiple cloud providers [20] It requiressometimes a complicated handling of Service Level Agreement (SLA) and responsibilityprinciples of interested sites [12] because in the general case the providers have to operatein an international context with different business regulations In fact they underlay todifferent legislatures in different countries Moreover they can be hierarchically organisedand be dependent on further international providers Therefore even by careful creationdeployment and maintenance of cloud services a lot of problems of multilateral data
82 4 Cloud Computing Virtualisation Storage and Networking
Table 42 Load balancing and functionality distribution between cloud computing and conven-tional IT (Representation by Microsoft)
Conventional IT IaaS PaaS SaaS
Applications + + Applications
Data + + Data
Runtime + Runtime Runtime
Middleware + Middleware Middleware
Web Services + Web Services Web Services
OS OS OS OS
Virtual Resources Virtual Resources Virtual Resources Virtual Resources
Server Server Server Server
Storage Storage Storage Storage
Network Network Network Network
+ For self-responsibility
Delivered from the cloud
security remain unsettled This factor limits in a certain kind of way the deployment rateand therefore also the advancement of the discussed new IT paradigm
Use of service technologies As cloud computing is essentially a set of service modelsmany of its issues can be understood when looking at how services are used and howcloud applications adhere to a SOA Such an architecture realised with web services inpractice (Fig 45) possesses the following benefitial advantages Web services offer loosecoupling and well-defined interfaces a good basis for EAI and application integrationacross organisational boundaries Furthermore they use open standards for protocols(eg HTTP) and content (eg XML or JSON) for which many development testing andusage tools exist so that new services can be consumed rapidly Using HTTP makes iteasy to produce and consume services according to the Representational State Transfer(REST) paradigm even though other protocols are also widespread Nevertheless thereare also weaknesses in service architectures which limit the full realisation of the cloudcomputing visions
1 Offering and consuming services dynamically asks for a service registry which servesas basis for selecting brokering and negotiating the terms of use The description ofservices within these registries is effort-intensive So far none of the effort distributions(by the broker by the providers by the crowd) has yielded a stable and completeregistry on a global scale
2 On a practical level an important complication is the configuration of security aspectsin deployed services Authentication authorisation access control and encryption arenecessary when leaving a closed trusted zone [5]
41 Clouds Technology Stack Basic Models and Services 83
Fig 45 SOAweb services basic architecture
3 The non-functional properties of services in particular QoS attributes need to bethoroughly defined and cross-checked at runtime As these specifications for m the basisof SLA documents a high-quality specification (ie high metaquality) inside servicedescriptions and a supporting environment with monitoring and adaptation support is anecessity
Some of the outlined problems can be solved or at least reduced with elaboratedextended web service specifications so-called RESTful services fully exploiting the HTTPspecification microservices and WS- The extended WS- use the basic components(Fig 45) and allow the creation of efficient service-oriented applications in various serviceenvironments including the web and in particular the ldquoSemantic Webrdquo The followingintegrated technologies and specifications are representatives for improvements [17 37]
1 Reliability via WS-Addressing WS-Reliability WS-Message Delivery2 Messaging via WS-Eventing WS-Notification3 Security via WS-Security WS-Trust WS-Privacy WS-Federation SAML (Security
Assertion Markup Language)4 Transaction Co-ordination Context via WS-Transactions WS-CAF (Composite
Application Framework)5 Semantic Features via OWL-S (Web Ontology Language for Web Services)
The extensions and their relations and layered placement are depicted in Fig 46 Basedon the REST model (Fig 47) the performance and scalability of services can be increasedby relying on an underlying HTTP server infrastructure These servers are typicallyhighly optimised and take care of caching streaming and other convenience functionalityRESTful web services act in some measure as an antagonism regarding to Simple Object
84 4 Cloud Computing Virtualisation Storage and Networking
Fig 46 Extensions WS- and alternatives
Fig 47 Representational state transfer method
Access Protocol (SOAP) and XML-RPC for which dedicated less common and lessoptimised server and client implementations need to be used
Such web services themselves and based on them further service-oriented and service-bound applications can be described according to the mentioned architectural style usingonly URIs as endpoint identifiers a contentresource model associated to each URI andHTTP in version 11 or 20 as interaction protocol The distinguishing features are asfollows asynchronous temporary character no RPC direct requests on resources anddocuments (URI) use of a generic interface standard semantics and stateless commu-nication protocol RESTful web services contain and convey the necessary context bythemselves and are operated only via simple methods (GET PUT POST DELETE) Suchsparingness leads to more consistency by the use of established standards On the otherhand a scalable a-priori analysis of the service features by description document analysisis not possible in this model Modern service description languages like Linked USDLand Swagger attempt to fill this gap Contemporary SOA concepts are mostly focused onEAI and B2B surroundings However the mapping of business processes (respectively
41 Clouds Technology Stack Basic Models and Services 85
for VTEO) as well as service orchestration and composition (eg via BPEL4WS) is stillinelastic and associated with higher developer-side complexity Therefore elaboration ofnew concepts is an imperative The concepts have to include not only new marketableideas eg like VTEO but also the analysis of costs and benefits [21]
Delegation of network functionality to cloud providers The functionality of a cloudis to deliver services by accessing the virtualised resources whose internal structure isunknown to the users providing certain common operations resource-intensive tasksconsolidation and distribution of resources and integration of applications in IT systemsof companies [23] Providers within an IoS deliver the services at different hierarchicallevels The functionality of the computers and further interaction devices as thin clients ofend users in the cloud is limited to providing a graphical or multi-modal interface (servicefrontend) caching the data selection of and access to external network services We seea resurrection of this host-node computing model in the increased use of consumption-oriented notebooks netbooks smartphones tablets and smart watches Access to networkresources can be provided by using the standardised web service protocols ExtensibleMessaging and Presence Protocol (XMPP) and SOAP including a range of extensionsto both for permanent sessions and request-response models respectively Access tothese resources can be also ensured via RESTful methods a session-less paradigm whichtransfers state by modifying resources on the server The processing and archiving tasksdatabase querying calling and encapsulation of further internal function calls are delegatedto the cloud provider There are closed (private) public and hybrid clouds which includefile servers databases archiving backup systems high-performance computers computergrids and multi-processor clusters Peer-to-peer clouds are not yet widely used but theyare considered as a future trend in research in particular for trustworthy mutual backupmainly driven by the exploded count of personal mobile devices SLA between cloudproviders and end users guarantee a certain QoS and aim to achieve a high level ofusersrsquo satisfaction called Quality of Experience (QoE) Cloud computing provides thefollowing functionality outsourcing of IT infrastructure to the cloud provider which maybe less expensive than maintaining a private one hosting of services saving costs foradministration and maintaining the IT infrastructure outsourcing of data archives andapplications (mail servers file servers databases backup services etc) cost-saving byusing high-performance computer clustergrids as a service
The main cloud models given by the NIST and Microsoft definitions have already beenpresented in Fig 43 They should be explained in greater detail and with examples SaaSis the model which directly appeals to end user It encompasses service-oriented webmobile or desktop applications (including virtual desktops) but also purely programmaticapplication and data services providing the access to resources in the cloud via thesediverse frontends PaaS provides an integrated platform for developing and testing webapplications (testbed) and eventually running them on a service platform with dynamicfeedback for the continuous development and advancement IaaS provides services ofvirtual networks by using remote servers systems of networked hard disc drives Virtual
86 4 Cloud Computing Virtualisation Storage and Networking
Machine (VM) with network management exploiting the SNMP protocol and upcomingOCCI interfaces The IaaS layer can be further subdivided into compute storage andcommunication resources
Example 41 CloudFoundry OpenShift and Bluemix are popular commercial PaaSplatforms There are very few non-commercial ones but there are a few prototypicalplatforms resulting from research projects including SPACE and FIWARE which mayinfluence future production platforms Vamp is an advanced PaaS server for complexservices whose implementation consists of orchestrated inter-dependent containers
Communication is an implicit prerequisite for compute and storage services so that theycan be used over the network For cloud backup systems the main interest is in storageresources which are accessed through network resources In practice these resources arenot universally described When creating commissioning and maintaining cloud servicesa lot of questions of IT security still remain open limiting the further spread of cloudtechnology This could be addressed by the creation of a non-profit cloud security allianceaiming to collect the best practices of effectiveness legal compliance and IT securityResearchers already started an outreach into this direction through surveys [1222] Theseabstract challenges shall now be demonstrated with examples from a selection of countrieswith a varying level of development and cloud adoption rates With regards to cloudcomputing legal acts of Ukraine regulate in general the operations in the area of IT securityand related fields (intellectual property telecommunications cyber-crime television) [6]They can be evaluated as systematic and complete regulation thanks to the considerationof existing international best practices One current scientific task is the optimisation ofthe service characteristics of these providers regarding QoS and QoE Great importanceis given to the uptake of mobile services based on LTE4G as well as future 5G networkswith access through modern mobile devices running on iOS Windows Phone 8 or AndroidOS and the newer challengers FirefoxOS Ubuntu Phone and Sailfish all equipped withweb browsers and personal data vaults The development of these technologies is widelysupported by governments of developed countries since it allows a significant resourcesaving but requires coordination of providers in areas of efficiency legal issues andIT security of clouds Hence for designing optimal cloud systems the non-functionalproperties of the physical hardware the network connections and the client integrationaround the software and services need to be considered and evaluated
Figure 48 highlights the relation between layered components of a cloud stackarchitecture and the resulting services which are offered for all of the layers
Cloud quality criteria It becomes evident that due to the high number of ofteninterchangeable services only through non-functional properties an automated distinctionbecomes possible These properties encompass primarily the quality (what do I get) andprice (what do I pay) properties The main quality criteria for cloud services are asfollows
41 Clouds Technology Stack Basic Models and Services 87
Fig 48 Context between cloud components and cloud services
bull Measurable QoS including execution performance response time and availabilitybull Comfort in use relating to the QoEbull Control by usersbull Reliability and data securitybull Price (per unit of data and time)
In Fig 49 a comparison of comfort vs control for certain well-known systems has beendone The evaluation was realised for the following systems Yahoo Facebook AmazonEC2 Salesforcecom Dropbox Google Docs in the organisation types of clouds hybridpublic and private
Hence to summarise while the consumption of cloud services is highly attractive itbrings along its own set of difficulties disadvantages and weaknesses in addition to theones inherent to general services
1 Performance and convenience of offered clouds are questionable and require actual useto find out
2 Lock-in to single vendors and cloud providers worsened by asymmetric pricing modelie uploading data is cheaper than downloading
3 Cloud providersrsquo creditworthiness trustworthiness and reputation4 Reliability issues or even total failure of providers (a provider can disappear from
horizon eg from economic legal or political reasons)5 Risks of temporary or permanent data losses or even leaks by providers
88 4 Cloud Computing Virtualisation Storage and Networking
Fig 49 Function comparison comfort vs control for certain well-known systems [11]
A concept of cloud-based virtual telecommunication office Among other trends thedevelopment of a modern VTEO based on SOA hosted in and delivered by a cloud isone of the up-to-date tasks and very profitable business niches We would like to dealwith a mentioned VTEO concept and certain significant examples and use cases [16]The world economics is widely characterised nowadays by the stable trends that thelarge and mid-range companies and authorities let in ever greater extent to outsource ownengineering services via external smaller service providers A concept for a modern virtualtelecommunication engineering office under use of SOA and cloud computing technologieshas been offered Multiple use cases for virtual telecommunication engineering office havebeen discussed As a significant example the CANDY Framework and Online Platformhave been examined The important development trends for the CAD for network planningregarding to the tool integration and effective access optimisation have been discussedThe CANDY system has been represented as an exhibit at CeBIT 2007 2008 2011 inHannover
The discussed service providers are as a rule independent highly-specialised engineer-ing offices acting with high-performance networks (VTEO) with relatively few employeesBut the mentioned VTEO systems can only survive in the long term if they provide theirservices at reasonable costs at the shortest time and on the highest quality level Letus refer to the offered services as Virtual Project Processing Examples of VirtualisedProcesses (VP) and the corresponding tasks circles can be formulated very largely Thereare inter alia the following tasks and processes electro-technical calculations chip andelectronic circuit design judiciary documents preparation statics computing for civil crafttax return bill preparation etc Accordingly the following specific requirements on suchVTEO systems have to be discussed in this section per client order (performed project)can be obtained a relatively high profit however its processing time is usually limited
41 Clouds Technology Stack Basic Models and Services 89
simultaneous processing of multiple projects in various steps of preparedness cooperation(via discussions and document exchanges) with several groups of clients delegationif necessary of the project steps (subtasks) to the partner agencies (ie subordinatedVTEO instances) participation of several specialists at each project efficient projectmanagement necessity of the exact project documentations at each processing steppermanent improvement of company Permanent improvement of companyrsquos know-howcan be effected via problem discussions successful qualifications and renewal training ofthe staff efficient knowledge storage reuse of project results in the subsequent projectsNowadays the current situation in most usual engineering offices is contradictive andcan be formulated as follows There is a highly qualified staff but also a very expensivestaff training use of modern CAD techniques (Computer-Aided Design) for individualengineering works (projects) but some inefficient cooperation of the participants hightime extensity and labor efforts for contacts to the client and partner companies
It is therefore an important scientific-technical problem to make the discussed tech-nologies available for VTEO With SOA (web services) and cloud computing techniques(private and hybrid clouds) aimed at an implementation of available services and providingaccess means are two indispensable components of the examined VTEO concept The mostacceptable models of the inter-operability scheme VTEO-2-Clouds are SaaS and PaaSFirst the VTEO must choose which kind of engineering services can be offered for therespective types of the projects and define for each an exact workflow of the project stepswith the subordinated tasks and the associated qualification requirements (specialist roles)At least one qualified employee has to be dedicated for each role For the individual worksthe high-quality CAD tools are to be provided as well as a powerful project managementsystem additionally for the project organisation aims It is important to provide that allproject documents are concurrently available for all the participants (specialists partnersclients) and they can efficiently communicate inter alia Furthermore the retrieving andon-demand offering the inter-operability of the most important project documents is tobe supported This requires specific document formats for each step of a project thatcan be processed in the subsequent steps without any further manual transformation Thediscussed concept of a VTEO is very helpful to meet the above mentioned requirementsThe resource requirements for such virtual engineering office move can be assumed tobe in the acceptable middle ranges (quantity of project employees amount of retrievedproject data) For general communication and collaboration means classical services canbe used (e-mail SSH Skype videoconferencing) The document management must becompletely centralised and web-driven For the access WWW techniques have to be usedpreferably (document preparation and supply per standard formats like HTML and PDF)For any special project data the appropriate XML-based professional problem-orientedlanguages are to be additionally developed with the associated XSDXSL (XML SchemaDefinitioneXtensible Stylesheet Language) The project workflow management is themost important part of the discussed virtual engineering offices But the majority of thecommercially available systems are anyway too complex for direct use Leaner solutionsare therefore preferable Such workflow management solutions are usually based on Gantt
90 4 Cloud Computing Virtualisation Storage and Networking
Fig 410 Project step 1 tasks 1ndash5 example execution period 1003 ndash 19032015 A typicalrepresentation of a workflow via Gantt diagram
diagrams (Fig 410) For each workflow step in a project there are the different processtypes Over and above that the following classification of process types for a VTEO canbe deployed automated with a simple communication scheme (without human assistanceand eg under support of sparing stateless protocol REST) half-automated with use ofcomplex stateful protocols with commits (under participation of specialists and dedicatedpersonal as well as under support of classical stateful SOAP over HTTP or other carrierprotocols) completely manual (expensive and very complex)
Purely human works (like eg granting of permission) have to be organised viathe WWW using web services web sites or mobile (web) applications Use of theworkflow management system is to provide the necessary download-functionality for inputdocuments and correspondingly after completion of the works (execution of businessprocess logic) the necessary upload functionality of the required resulting documents bythe responsible project employee to the centralised document management system Theworks with the CAD tools like eg ArchiCAD are to understand as defined above asthe purely manual works It is particularly efficient if the VTEO can offer a processingsupport also via a central platform This can be realised especially efficient on AJAXbased techniques The user activities are executed within the standard WWW browsersthe business logic processing follows at the server site eg via activation of certainspecialised scripts The resulting documents will be stored automatically and project-specific at the server site The specific workflow-centric management for a VTEO mustbe defined under use of the following principles and requirements to the process elementsand their synchronisation a workflow is combined from a sequence of design stepseach step consists of one process (task) or multiple parallel processes each processpossesses a status eg (ready (yn) result (+-)) each process uses andor producesinputoutput documents a process is either an atomic process or a workflow by itself
41 Clouds Technology Stack Basic Models and Services 91
The next important aspect is a type of billing and a payment method (accounting in aVTEO) There are different possible systems between the simplest blanket (all-in-one)accounting of delivered services to differentiated complexes prices depending on dataamounts manual efforts tasks dimensions and computational complexity With the simpleVTEO accounting forms SSL method or alternatively XML security find favor SETmethod can be recommended for differentiated complexes prices schemes The discussedissues are illustrated sufficiently in the next sections of the given work on the example ofa VTEO (a fictive service provider) for a design of combined network structures
Conclusions and research fields regarding the clouds The most important tasksoriented at the elaboration of the advanced clouds that are free of the above-mentioneddisadvantages can be listed [13] They are grouped into three groups
bull Cloud adaption and optimisationbull Strategies for the compensation of SLA violationsbull Strategies for minimisation of energy consumptionbull Mechanisms for the visualisation of complex cloud monitoring databull Deployment of RAIC with cockpit features at the customer sidebull Fine-grained SLAbull Methods to determine fine-grained properties of cloud servicesbull Identification of assets and corresponding requirementsbull Deduction of monitoring targets from SLAbull Cloud surveillance and incident detectionbull Specification of monitoring targets and SLA violationsbull Models for the proactive recognition of SLA violations and the evaluation of a cloudrsquos
energy efficiencybull Mechanisms for reliable distributed monitoringbull Dynamic provider selection and cloud setupbull Flexible distribution mechanisms for cloud platformsbull Strategies for the performance optimisation of cloud applicationsbull Reputation consideration to improve reliability and trustworthiness
An example of an advanced cloud technology with transparent encryption is illustrated viaFig 411 The features of the transparent encryption are as follows
bull Safe hybrid access D public C privatebull Efficient cryptosystems AES RSA MDMACbull Analysis of structured and unstructured databull Document classification and codecs demarcationbull User authentication and key distributionbull PKI deployment with the certificates (X509Kerberos)
92 4 Cloud Computing Virtualisation Storage and Networking
Fig 411 MD ndash Message Digest MAC ndash Message Identification Code AES ndash Advanced Encryp-tion Standard RSA ndash Rivest Shamir and Adleman Encryption PKI ndash Public Key Infrastructure(X509 Kerberos) Secured cloud with own controller [11]
42 Virtualisation of Services and Resources
Nowadays a virtualisation of services and resources is required due to the heterogeneoushardware and applications landscape and the increasing overcapacity in single devices(Figs 412 and 413) Virtualisation methods became wide-spread since 1990 and offernow a necessary entry or preliminary stage to the modern clouds
The statistics 2014ndash2015 demonstrated an approximated distribution for landscapediversity of applications and apps (Fig 413) Among them are regular desktop appli-cations SaaS (thin) clients within clouds mobile applictions as well as usual webapplications under a variety of operating systems
A classification system examples as well as advantages and disadvantages arediscussed below A useful classification of virtualisation methods is given in Fig 414Certain of the listed methods for the virtualisation of services and resources can be orderedto different classification criteria (hardware software applications server containernetwork) as well as the evolving SDN
The basic virtualisation unit for compute resources is the so-called VM which offersa single service a complete operating system or an application Efficient deployment andmigration of VM is controlled with different methods The most important of them are OScontainers hypervisors and VMMs (VM monitors) A layered architecture with 3ndash5 layers(HW OPS virtualisation layer etc) is a typical construct The comparison between OScontainers hypervisors and VMMs is given in Fig 415
A typical solution for UNIX-like operating systems is so-called spartan BSD jailswhich exist in similar form on Linux (chroot) and on Solaris (zones) They are practicallydedicated to a single specific application but in principle allow a complete interactive
42 Virtualisation of Services and Resources 93
Fig 412 Motivation heterogeneous hardware
Fig 413 Motivation heterogeneous applications landscape
session with sub-processes The disadvantage of the jails is located in their near-absoluteisolation Citrix-based solutions offer a mostly comfortable virtualisation concept withmonitoring of VM without host OS as additional layer of virtualisation [2] The Hypervisoracts as a meta-OS VMware products use as a rule a VMM pure to control VM which aredeployed over the host OS Hypervisors and VMMs offer a lot of advantages in comparisonto the containers except the highly-secured runtime environment An example would be asandboxing container within a mobile OS with foreclosed apps under reputation codeand antivirus control The mostly used types of hypervisors are depicted in Fig 416A frequent use case is the virtualisation of previously dedicated hardware servers for ratherlight-weight functionality (e-mail domain file storage or backup)
94 4 Cloud Computing Virtualisation Storage and Networking
Fig 414 HW ndash hardware OS ndash Operating System NW ndash network VM VMM ndash VM-MonitorSDN ndash Software-Defined Networking Classification of virtualisation methods (Own review)
Fig 415 Classification of virtualisation methods
42 Virtualisation of Services and Resources 95
Fig 416 Certain types of hypervisors
Fig 417 An example of virtualisation
Example 42 In Fig 417 an example herewith is depicted The specified VM and VMMsenable a flexible and efficient solution for web presentation consumer portal as well aslegacy software
bull each VM is an independent isolated from other VM platform for any guest OSbull VM can behave as it would possess the host computer alone (but insignificant
slowdown)bull in the desktop area mainly tests or simulation environments were performedbull VMM concept is widely used to increase the utilisation and availability of servers and
reduce the costs (procurement maintenance personal power HVAC) as well as ROI
96 4 Cloud Computing Virtualisation Storage and Networking
Fig 418 VMware layered architecture own review based on IBH Dresden Professional IT-Services (Source ibhde)
Fig 419 VMware Horizon Suite product features
The major products on the market offer a complex layered architecture like in Fig 418The depicted architecture is typical for VMware products
Many virtualisation solutions offer a central management console to orchestrate alltasks The product features for VMware Horizon Suite are given below (refer Fig 419)
42 Virtualisation of Services and Resources 97
Fig 420 The Citrix products on BYOD
A Citrix platform for mobile collaborators as well as flexible mobilewireless platformfor the known BYOD problematics (ldquoBring Your Own Devicerdquo) with application virtuali-sation concepts is depicted in Fig 420
Example 43 A company with a heterogeneous computing environments is about tovirtualise their IT hardware How is the data exchange between such heterogeneouscomputer systems realised In a company network with 30 computers there are 3 differentarchitectures (Fig 421)
(a) How many importexport routines must be programmed and installed for interoper-ability (understanding) between all systems is possible (b) What changes occur whenanother 31st computer with novel system architecture is integrated into the network(c) What are the advantages and disadvantages compared to (b) as a result if thevirtualisation concepts are used
Virtualisation advantages from a business perspective Virtualisation is not only atechnical method On a strategic or financial level if when and how to virtualise is animportant decision process There are the following virtualisation advantages from pointof view of a company
1 Different virtualisation techniques are used for the areas like banking e-commercecivic craft financing assurances building society savings and trust companies Theyare a preliminary stage for cloud computing
2 The significant advantage of resource virtualisation is significant for CAPEX andOPEX (cost reduction) for SMEs and large companies
98 4 Cloud Computing Virtualisation Storage and Networking
Fig 421 Heterogeneous environments with virtualisation in a company
3 The large financial institutions obtained virtualisation solutions which displace allbefore processed transactions and applicationsrsquo infrastructures under use of sole serversand old mainframes
4 Virtualisation allows the operation of several available VM on a host5 Virtual servers provide virtual OS and runtime environments using VM in order to
maintain existing software (legacy systems) and allow use of mobile apps6 Virtualisation retains the heterogeneity of the network (SDN) and runtime environments
and hides from diversity of implementation details and restrictions in common OS andsoftware
7 Virtual servers can increase efficiency of operational IT infrastructure their utilisationand availability
8 Advanced EAI and B2B for corporate applications as well as for inter-company systemsby EDI and e-business (middleware SOA)
Example 44 What is VMware virtualisation today The distinguishing features of thisvirtualisation product are as follows (Fig 422)
bull Virtualised guest OS Windows Linux Mac OS X Chrome OS and othersbull Secure data access and deployment of apps and databull Work from anywhere deploy and migrate VMbull Optimise the network traffic backup and VM snapshotsbull Secure surfing within the clouds
42 Virtualisation of Services and Resources 99
Fig 422 VMware What is the virtualisation with VMware nowadays
More recently virtualisation of individual compute resources has evolved into an inte-grated data centre concept A software-defined data centre offers the following advan-tages
bull agilitybull controlbull efficiencybull freedom of choice
Virtualisation and cloud stacks can be run in co-operation as shown in the example withVMware RSA Security EMC2 and OpenStack
bull Public private cloudsbull United managementbull VIO concept VMware OpenStack (Fig 423)
Virtualisation with VMware implies the following
bull VMWare Data Protectionbull VMWare VSAN Architecturebull VC = VCenter Server v60 (Table 43)
The advantages are as follows
bull proactive IT availabilitybull innovation and dynamics
100 4 Cloud Computing Virtualisation Storage and Networking
Fig 423 VMware architecture
Table 43 Properties ofVCenter Server v60
Structure units Windows Linux
Hosts per VC 1000 1000
VM per VC 10000 10000
Hosts per Cluster 64 64
VM per Cluster 6000 6000
bull security and mobilitybull market chances by know-how insufficiency or limited resourcesbull attractive costsbull no fragmented datacomputing centersbull growth in equipment
An example with the dedicated hardware for VMware
bull EVO RAILbull Hyper-convergedbull Infrastructurebull Appliance
42 Virtualisation of Services and Resources 101
Proactive IT What does it mean to manage proactive IT via VMware
bull fast developmentbull providing of all applicationsbull optimised for each end devicebull Data center virtualisation and hybrid cloud extensibilitybull Native security controls in the infrastructurebull Optimised and automated data center operationbull Automation of infrastructure and application deploymentbull High availability and stable infrastructure
Each application everywhere is one of the mantras of virtualisation product vendorsFurther mantras are open management and united platform These slogans will be outlinedbriefly now The everywhere mantra leads to the development deployment and executionof convenient and modern applications The open management refers to the flexibilityto manage cloud infrastructure and applications Finally the united platform connectsinternal and external clouds with a common software-defined data centre platform basedon virtualisation concepts In the case of VMware the vendor calls the solution a hyper-converged infrastructure
Not only compute resources but also storage resources benefit from virtuali-sation concepts The VMWare mixed backup is based on the rotatory principle(Fig 424)
bull Full Backupbull Incremental Backupbull Synthetic Backup
The VMware cloud platform thus combines the following characteristics
bull management of all public cloudsbull VMware vRealize Suite for management of multiple public and private clouds (cloud
cockpit)bull optimisation of OpenStackbull VMware Integrated OpenStack (VIO) for the flexible and reliable entrance in the
OpenStack cloud of enterprise classbull integration in container toolsbull aimed at fast development and supply of new native cloud applications
VMware vCloud Air is an add-on product for virtalised desktops which provides thefollowing vitalisation startup help
102 4 Cloud Computing Virtualisation Storage and Networking
Fig 424 VMware mixed backup
bull desktops hosted in the cloud and available on demandbull increased user productivity and optimised IT operationsbull extension of existing applicationsbull 100 compatible the same security high availabilitybull web and mobile applicationsbull faster development of web and mobile applicationsbull vCloud Air development testbull 100 compatible lower cost broad OS support high availabilitybull disaster recoverybull simple cost-effective failover and restore
Example 45 The company Veeam has been founded in 2006 in Switzerland possesses2000 collaborators and serves 170000 users The hybrid virtualisation platform of Veeamis based on the software from Citrix VMware and Microsoft Hyper-V [7] The productsfor the hybrid virtualisation platform of Veeam are as follows (Fig 425)
bull ONEbull management pack
42 Virtualisation of Services and Resources 103
Fig 425 Hybrid virtualisation platform with Veeam
bull backup amp replicationbull explorer for storage snapshots
The architecture of the Veeam backup storage integration is shown in Fig 426 Thefollowing storages and products can be used HP StoreOnceCatalyst Support EMCDataDomainBoost VM Backup-File Chain HP StoreVirtual 3PAR NetApp ONTAP aswell EMC The essential advantages of this platform include the ability to support analways-on business ad-hoc restores of virtual machines as well as automated verificationof the state of virtualised applications Virtual machines can be instantiated and activatedquickly from both ISO images and snapshots from previous execution runs
The procedure of efficient backup based on snapshots with Veeam is depicted inFig 427 The creation of snapshots by Veeam for the backup is up to 15 times fasterthan the pure backup The Veeam Explorer for storage snapshots provides the backup ofthe following data items either all VMs completely or only guest files or all directoriesor specific folders of applications such as Sharepoint and Exchange folders
A mixed backup (consisting of differential + incremental runs) is provided Theexperimental 3-2-1-0 rule is valid in this case It refers to 3 media types for retrieving2 diverse backups 1 always available and 0 problems with it
104 4 Cloud Computing Virtualisation Storage and Networking
Fig 426 Backup storage integration with Veeam
Fig 427 Efficient backup of snapshots with Veeam
An example of backup frequencies for the following 4 years is depicted in Fig 428It differentiates weekly (4) monthly (12) and yearly (3) cartridges The standard LTO-Ultrium streamers and band cartridges can be used with Veeam in such scenarios
43 SDN ndash Software-Defined Networking 105
Fig 428 Example of backup frequencies wit Veeam
43 SDN ndash Software-Defined Networking
Virtualisation of network resources and software-defined networking Software-configured or defined networks are called SDN This term expresses a virtualised layerednetwork for data transmission in which the management plane of the network is separatedfrom the data transfer devices and has to be implemented programmatically SDN is oneof the known forms of virtualisation of computing and networking resources includingnetwork services and applications Its origins are in the backbone networks of telecomoperators but some of the mechanisms are now appearing for centralised configurationof multiple consumer devices as well The basic principles of future SDN developmentand deployment have been formulated in 2005ndash2006 by researchers from Berkeley andStanford universities even though the topic gained prominence quickly by heavy industryinvolvement
SDN motivation The main problem in the modern and very performant physicalnetworks is as follows
1 The traditional physical networks are heterogeneous too static for modern businessapplications and cloud services
2 Deployment virtualisation technologies are required3 Nowadays the applications are distributed between multiple VM that communicate
intensively With the goal to optimise workload of the servers VM instances oftenmigrate and hence change the ldquobinding pointsrdquo for the network traffic
4 Conventional addressing schemes logical dividing into VLANs and the appointmentof traffic rules in such dynamic environments become very ineffective
106 4 Cloud Computing Virtualisation Storage and Networking
Fig 429 (a) No virtualisation (b) SDN general architecture Motivation to software-definednetworking
5 As networking protocols evolve the firmware on networking equipment such asswitches and routers needs dynamic updates in a controlled and consistent manner tothe extent that it must be completely implemented in software
SDN solution approach SDN can be classified as the part of the network virtualisationSDN is per definition a resource virtualisation type like OS server or applicationvirtualisation (Fig 429 refer the classification in Fig 414) Simultaneously SDN is anapproach to the construction of computer network equipment and software where thetwo main components of such equipment are abstracted from each other via (1) controlplane (2) data plane and as a rule with (3) a protocol named OpenFlow to combineand coordinate L2L3 networks via VM deployment [15] Starting around 2013 SDNwere widely deployed by multiple manufacturers inter alia VMware Juniper BrocadeCisco HP and IBM By that time it became one of the main innovation topics along withcloud computing and big data with similar confusion about the technical depth and thehype portions of the innovation Let us discuss its advantages SDN enables a networkadministrators to perform simpler low-level management of the networks by abstractioninto virtual services SDN offer (refer Fig 429)
bull emulation of MAC frames and packets (MPLS IP LAN mobile radio) on L2 and L3bull deployment of zones user demarcationsbull cloud services in multi-tenancy agreementsbull diversity of SDN architectures via the availability of multiple providers
Refer to Fig 429 just once more and compare (a) and (b)One of the driving forces for the large installation base of SDN networks is a universal
protocol called OpenFlow which is independent of the manufacturer and implements theinterfaces between the logic controller for the network and the network transport A typical
43 SDN ndash Software-Defined Networking 107
Fig 430 A typical flow chart in a network device that supports the OpenFlow protocol
traffic table within a network device that supports the universal protocol OpenFlow isshown in Fig 430 With the use of OpenFlow a more flexible and efficient physical(MAC-) and logical (IP-) addressing becomes possible as well as the reconfigurationis supported for data flows services applications and application ports The OpenFlowprotocol provides traffic identification by using the term ldquoflowrdquo A flow table acts as a keyelement of a switch that supports this protocol similar to a rule table within a softwarepacket filter The group of columns on the left side of the table creates the matchingfields where the characteristics of the flow are represented There are different parametersincluding MAC and IP-addresses of the sender and recipient VLAN identifier TCP andUDP ports and other information These data entries are recorded via the controller underuse of the OpenFlow protocol and registered into the switch table (refer Fig 430)
Example 46 Due to the inset of a new VM the reconfiguration process for all accesscontrol lists on all network devices and levels in a large network may take several daysinto account The reason is that the orientation of existing management tools to work withsome concrete devices at best purpose offer automation parameters which apply to a groupof devices which belong to the model row of one particular manufacturer eg Cisco MIBIn particular the well-known system VMWare provides (Fig 431) the following softwareand services for SDN and its virtual devices [7]
bull Network access to SDN is determinedbull Use of physical plants in the networkbull Deployment of multiple VMbull Deployment multiple Layer 2 VLANsbull Inset of so called Virtual Distributed Switches (vDS)
108 4 Cloud Computing Virtualisation Storage and Networking
Fig 431 VMware-based scenario with access demarcation within SDN
bull Use of virtual network cards (vNIC)bull Use of VPN (Virtual Private Networks) and Load Balancersbull Deployment of network devices with proprietary VXLAN (Virtual Extensible LAN)
protocol that supports SDN within VMWare products as the alternative to OpenFlowbull A special system vNCS (VMware vCloud Network and Security)
The product palette of VMware is deployed VLAN SDN safety zones The networkinterfaces vNICs are coupled to dedicated virtual switches vDS that enable the distributionof VM assigned to the port groups of vDS Each vDS is not closely assigned to a servernext to each other but is configured to several servers Access demarcation within SDN isorganised with use of vSwitches
The network adapters of the servers are coupled to the vDS and allows VM on portgroups on the vDS the connection to the network This vDS is not tied to a particular serverbut is configured across multiple servers Use of vShield Zones is as follows virtual datacenter enables basic VM-protection against network threats (firewall packet filtering) Thesoftware vNCS (VMware vCloud Network and Security) is used with the aim
bull Deployment of a specialised VXLAN protocol (Virtual Extensible LAN)bull Deployment of virtual firewallsVPNs load balancing elements (load balancers refer
to the picture)
43 SDN ndash Software-Defined Networking 109
Fig 432 Deployment of vSwitches
The implementation of the principles of SDN using virtual switches of the type vSwitch isdepicted in Fig 432 The mentioned decision on virtual switches of type vSwitch level L2has many options including devices by VMWare Juniper Cisco HP and IBM for accessvia the level L3 gateway (GW) to the virtual machines with specific applications networkservices and cloud services Available data protection against malware and many possibletypes of threats on the network layers L2 L3 L4 L5ndash7 is achieved through the use offirewalls and antivirus software (see Fig 431)
SDN evaluation The features of SDN are presented in this section SDN provides theefficient separation of traffic transmission functions in few layers
Use of SDN offers evident advantages Routine network reconfiguration functionsare so simplified that the administrators do not have to separately enter hundreds ofconfiguration code lines for different switches or routers The network parameters canbe also changed quickly even in real time thanks to a rapid propagation of the parametersand rules Accordingly the timing of the introduction of new applications and serviceswill be greatly reduced The SDN technology uses expediency and efficiency in futuregeneration of mobile communication 5G by the defining IMT 2020 standard SDN willbe part of the future 5G mobile connections Together with 5G a number of terms havebeen declared which may express future innovation or further hype topics Examples arethe intelligent web of connected things real-time remote control mobile cloud trafficimmersive experience lifelike media ubiquitous connectivity and telepresence Moredetails about the aims of 5G networks are provided in chap 6 Software implementations of
110 4 Cloud Computing Virtualisation Storage and Networking
a prototype for a provider core network according to 5G may be based on networks usingprotocols of SDN like OpenFlow VXLAN and virtualised operating systems based onVMWarevSwitch Citrix products and similar ones SDN are effective for the constructionof the cloud services infrastructure in conditions when by a request from users it isnecessary to create a virtual node a virtual service automatically and quickly Herewiththe virtual network has to allocate the required resources autonomously As a part of the5G mobile generation 5GIMT 2020 SDN technology becomes feasible in large datacenters allowing to reduce support costs by centralising network management as wellas by increasing the usage of network resources through their dynamic managementUse of SDN in practice will happen primarily for provider cores including 5G mobilenetworks to allow the telecommunication carriers and independent providers to obtain thenew management functions and better control via network components and services of anytype from a single centralised location which will greatly simplify their operation
44 Backup Services within Clouds as Advanced Cloud BackupTechnology
Next to virtualised compute and networking resources storage resource services are alsopopular in many cloud applications There are multiple flavours including higher-leveldatabase services file services and low-level block devices offered as service on which acustom file system can be placed The following text concentrates on file services as thisis the flavour most commonly used in consumer applications
Data crashes can cause unpredictable and even hard-out effects for an enterprise orauthority Backup strategies as antidote unify a complex of organisational and technicalmeasures that are necessary for data restoring processing and transfer as well as for datasecurity and defence against its loss crash and tampering [4] High-performance modernInternet allows delivery of backup functions and is complemented by attractive (mobile)services with a QoS comparable to that in Local Area Networks One of the most efficientbackup strategies is the delegation of this functionality to an external provider an onlineor cloud storage system This article argues for a consideration of intelligently distributedbackup over multiple storage providers in addition to the use of local resources Someexamples of cloud storage deployment in the USA the European Union as well as inUkraine and the Russian Federation are introduced to identify the benefits and challengesof distributed backup with cloud storage
Motivation Up-to-date network technologies aimed at backup and restore technologiesof critical enterpriseauthority data are discussed A comparative analysis of existingcomplex solutions and standalone tools is represented Essential advantages in restoretechnologies for critical enterprise or authority data can be offered via a newly devel-oped original cloud backup concepts in comparison with the traditional data-centricbackups But the complex constellation of international law and multilateral data safety
44 Backup Services within Clouds as Advanced Cloud BackupTechnology 111
requirements limits in some way the development of network technologies for cloudbackup One of the possible ways for solving the mentioned problems is offered byan intelligent combination of well-known commercial storage clouds with the use ofefficient cryptographic methods and stripesparity dispersal functionality for authenticatedtransparently encrypted and reliable data backups This approach has become popularrecently under the name RAIC [10 29 33] Yet from both a scientific and a practicalperspective there are shortcomings in conventional RAICs when eg dismissing the costand trust characteristics of the associated storage services
441 Backup as Important Component of Informational Safety
Disruption of critical data has unforeseen and heavy consequences for companies ororganisations It may have different reasons but the main result remains always the samea significant risk of losing data or access to it This may lead to impediments in reachingthe goals of companies or organisations errors in documents malfunctions of tools andmachines losing reputation on the side of partners Very often the risks of losing data arecaused by natural phenomena as shown in Table 44 where they are presented along withstatistical probabilities and human factors
The next problems of the company or organisation are significant costs for the recoveryof critical data and compensation of damages For these reasons backup technologies area very practical task and a relevant part of securing data and assuring information safetyof the company or organisation The purpose of data backup is the regular creation ofcopies of files databases applications and settings on external backup systems whichin most cases are storage units managed by a backup application Modern networkoff-site backup systems support this process with separation of locality for reasons of savingand recovering the data and prevent the risks of data loss in a company or organisationthat may appear because of hardware malfunction due to voltage jumps or devastating
Table 44 Causes andprobabilities of losing criticaldata due to natural and humanfactors
Cause of losing data Statistical probability
Natural phenomena
Hurricanes 1
Fire 6
Water 8
Short-circuit 16
Lightning stroke 17
Other natural phenomena 17
Human factor
Usage faults 25
Stealing 10
112 4 Cloud Computing Virtualisation Storage and Networking
Fig 433 Example of backup system structure
natural disasters such as fire water attacks of malicious software like computer virusesand trojans system errors during data storage stealing the data or accidental dataleaks Backup includes organisational and technical measures for storing processing andtransferring back important data and guarantees their protection from loss destruction ordisruption The main distinctive features of modern network backup systems are the targetdevices (smartphone tablet PC rack server form factors) along with the target storagemedia (magnetic disks or tapes electronic flash memory and optical disks) delay of dataaccess (in the ms range up to several min for cold backup) maximal time of safe datastorage (months years) error rate GB costs An example of a combined backup systemfor a small or medium-sized company or organisation is shown in Fig 433
The main components of the system are an optical network (ATM 10GbE) SAN atape library and Redundant Array of Independent Disks (RAID) file server systemsAccording to Table 45 the main criteria for the choice of suitable backup media andnetworking technologies include high-speed connections (1 GBs over LAN) very largedata volumes of overall storage (from 100 Petabytes up to Exabytes) long guaranteedusage duration (months years) all when at the same time having a low probability oferrors and costs per data unit This list is not conclusive good handling of small files andbackup schemes are further factors
As it can be seen from Table 45 the streamer tools (Streamers SLR DLT DATDDSLTO VXA) guarantee a low probability of errors and costs per data unit long guaranteedduration and large data volumes as well as a good pricevalue ratio But a non-linearrestore operation from such media is a time-consuming task leading to the requirementof balanced choices The RAID mechanism is based on the creation of a redundant arrayof independent (multiple vendors) and inexpensive (consumer SATA instead of SAS) harddisc drives (HDDs) which work in one system to improve selectively both speed and
44 Backup Services within Clouds as Advanced Cloud BackupTechnology 113
Table 45 Overview of backup media
Media for backupMax datavolume Cost per 1 GB
Guaranteedusage duration
Probability offailures
DVD 47ndash85 GB 005 Small 1 year High
USB flash 2ndash256 GB 097 Very small Medium
USB-HDD 05ndash4 TB 004 Very small Medium
Streamer LTO 02ndash3 TB 006 30 years Low
Streamer DLT 016ndash16 TB 017 30 years Low
Systems of redun-dant discs RAIC
Max 10 TB Multiple ofHDD costs
Several years Low
reliability of IO operations The array of HDDs is controlled by a special RAID controller(hardware or software array controller) which provides the functionality of storing andretrieving data in the array as well as creating and checking the checksums This allowsmaking the underlying system transparent to the external users and presenting it as onelogical IO channel Thanks to parallel runs of readwrite operations on several discs thedisc array provides a higher speed of data exchange compared to one large disc
The RAID mechanism was created first in 1988 by D A Patterson G Gibson and RH Katz researchers of University of California Berkeley For regular backups differentvariants of underlying storage types exist streamers connected via local network (method1) backup via LAN (method 2) backup via SAN (method 3) backup via NAS (method4) backup via external backup provider (data center or cloud system) (method 5) Foroccasional backups removable media such as USB sticks and portable hard drives mayalso be an option But due to the criticality of backup this is one of the processes whichreally should be automated
For choosing the right backup method for a company or organisation different methodsand factors should be considered including size of the company or organisation structureof available networks number of users (a small enterprise with 20 users or a big companywith more than 1000 users) costs of backup requirements on data safety and security aswell as administration efforts In recent years network technologies made a great progressin QoS (due to WdM 10GbE) mobility (HSDPA LTE) and easy access to computingcenters In fact the emerging IoS ensures that application based on SOA principles havebeen created which naturally integrate into service environments and can discover anduse suitable backup services without manual configuration High-speed Internet enablesproviding functionality and services with the same quality as known from local networksand hence makes the shift of formerly relatively local functions such as backup into thenetwork feasible The new IT paradigm of delegating the services to external providersis known as cloud computing and when referring to backup as cloud storage One ofthe most effective backup strategies is thus the delegation of the entire backup processto an external provider by interfacing with up-to-date cloud systems This is achieved byplacing the backup services into a public cloud offered by a capable and trustworthy cloud
114 4 Cloud Computing Virtualisation Storage and Networking
provider Cloud computing is becoming more and more popular when several companiestransfer their IT infrastructure (completely or partly) into clouds This may lead to a lackof transparency of data access (who when where why and what) and cloud reliabilityand raises the risk of loss of all critical data if the cloud provider leaves the market Tomitigate these risks to some extent the deployment model of private clouds (method 6)under operational control from the client may be used Furthermore intelligent client-sidetechniques can further reduce the risks Below a very precise definition adopted from theNIST and Amazon definitions of the concept of cloud computing is given [1 24] ldquoCloudcomputing is the on-demand and pay-per-use application of virtualised IT services overthe Internet The clouds can offer on-demand self-service broadband network accessresource pooling measured and optimised service rapid elasticityrdquo The adoption ofcloud computing provides the following advantages relative reliability and security whilegiving up physical possession staying in control when demand changes the controlcan be exerted through vertical and horizontal scaling and migration to other providersavailability of attractive multi-layer services from infrastructure to software applicationsefficient platformsstacks and convenient client integration (Table 46) The broad range ofplatforms and choices in functionality leads to a discussion of the most important domain-specific criteria for cloud backup These criteria based on those for general backup andthose for general cloud computing are QoS parameters such as throughput data ratedelays and reaction time convenience (comfort suitability effectiveness) user controltrustworthiness security and privacy price per data extent and time
The next position might be the organisational reliability (trustworthiness of a cloudprovider) because a provider can disappear from the horizon unexpectedly for instancedue to own economic legal or political reasons Data security is required since therisks of data losses and compromises by provider maintenance via third parties are stillunreasonably high
Regular backup software Backup software is the basis for the realisation of any backupstrategy in a company or organisation which allows the automation of the backup tasksThe software triggers the backup process in a certain point of time provides the fullor incremental backup of the selected data and arranges for an appropriate reportingto inform the IT administrator among other goals The software may run in push modeas scheduled software application on each device or VM to be backed up or in pullmode where agents are connected to a backup service The choice of backup softwareand services may include fully extensible open source software as well as proprietarysoftware which has limited configuration and customisation options In both cases theoffer may be for free or based on a purchase or subscription contract to include supportGenerally the choice for a backup software depends on the required functionality transfereffectiveness restore performance and reliability The commercial solutions may howeverlead to a backup software and service lock-in which should be avoided similar to a storageprovider lock-in This is why in all backup planning projects a compromise should be made
44 Backup Services within Clouds as Advanced Cloud BackupTechnology 115
Table 46 Well-known cloud platforms
Platform Provider
Amazon EC2 Amazon Web Services (AWS) for Elastic Compute Cloud(EC2)
Cloud Computing Yahoo Cloud services from Yahoo Platforms
Cloud Computing Resource Kit Cloud services from OracleSun
Eucalyptus IaaS stack which reimplements the Amazon APIs
SalesForce Cloud services from Forcecom mostly on the SaaS level
Google App Engine Google (a PaaS model)
Google Docs Google (a SaaS model)
Google Compute Engine Google (an IaaS model)
iCloud A virtual OS on a Cloud basis
Meebox Online file management in the frame of a SaaS model
MS Windows Azure Multiple Cloud Services in the frame of the Win Azure Platform(Microsoft)
Nimbula A privatehybrid cloud technology of former AWS-collaborators
OnLive An interactive Games-on-Demand-Platform with compressionmethods for computer graphics and videogames
Open Cirrus Open Cloud Computing Research Testbed from opencirrusorg
OpenStackorg Open Cloud from Rackspace Citrix NASA Dell
OpenNebula Commercialised European research project for data center vir-tualisation and service markets
OpenShift PaaS from Red Hat
T-Systems Dynamic Services A private Cloud-system for dynamic deployment of SAP-applications from SAP GmbH
Verpura Online-Cloud for Enterprise Resource Planning in SME
VMware vSphere A virtual OS on the Cloud-Basis of VMWare
between the costs and added value of the backup solution (functionality effectiveness andreliability) cf Table 47
Modern systems for cloud backup One of the most promising backup strategies is todelegate backup to an external provider eg to a cloud backup system A short overviewof cloud storage providers suitable for backup is given in Table 48 Online cloud resourcebrokers and marketplaces are updated periodically for an up-to-date view on the choicesbased on rich provider descriptions which facilitate the exchange of the informationthrough open markets A comfortable access to the cloud backup systems is possiblethrough dynamic and non-intrusive service selection even with mobile devices like tabletsor smartphones If the company or organisation does not trust the cloud provider it coulduse the technology of private clouds which limits the access to the cloud for external users
116 4 Cloud Computing Virtualisation Storage and Networking
Table 47 Selected backup software
Software Description Costs
DAR (Disk Archive) Uses an own archive compression format dis-tributes the backup copies into different frag-ments and discs supports common encryptionmethods
Freeware
Rsnapshot Creates hard links between different storedroutes that requires the storage media support ofthe hard links When a file changes not only thechange difference is backed up but the wholefile
Freeware
Duplicity Creates backup copies in encrypted formatGPG (PGP) and archived in GZIP Backupcopies can be made practically for all types ofoperation systems supports upload of backupcopies over FTP systems SSG Rsync Web-DAV HSi and Amazon S3
Freeware
Acronis Backup ampRecovery AdvancedServer
Popular but expensive software for MS Win-dows allows creating image and file backupsis oriented on using HDD tape libraries cloudtechnologies
About 1100
Drive Backup Server Provide different backup functions eg storageon internal and external media CDDVDBRdiscs NAS systems FTP with support of virtualmachines VMWare
About 500
Symantec Backup Exec2012
Similar to Drive Backup Server About 900
Rsync Allows scripts for configuration of shell copy-ing files and their parts The special feature ofRsync is effective synchronisation of file treeover network
GNU GeneralPublic LicenseUnix-Distributions
Cron-Daemon System process of Unix for timer-based trig-gering of processes like backup The backuptasks can be triggered periodically accordingto ldquocrontabsrdquo tables and are called ldquocronjobsrdquoThey create backups on specified servers
Unix-Distributions
Bup A combination of Rsync and Git (version con-trol) concepts It offers Par2 redundancy
GNU LGPL v2
Bacula Client-server based network backup applicationfor individual computers up to large networks
GNU AGPL v3
Amanda Advanced Marayland Automatic Network DiscArchiver with support for tape drives disks andoptical media with native Windows client
BSD-style
44 Backup Services within Clouds as Advanced Cloud BackupTechnology 117
Table 48 Overview of cloud backup platforms
Name of cloudbackup system
Region ofstorage
Max volumeof cost-freestorage
Max volumeof paidstorage Platform
Amazon CloudDrive
USA 5 GB No limits Win Mac Linux iOSAndroid WindowsPhone
Dropbox USA 2 GB No limits Win Mac Linux iOSAndroid Blackberry
Windows LiveSkydrive
Ireland 25 GB 100 GB Win Mac WindowsPhone iOS Android
Strato HiDrive Germany ndash 5000 GB Win Mac AndroidWP7 Chrome Synology
Google Drive USA 5 GB 16000 GB Win Mac iOS AndroidLinux
HighSecurityBackup
Germany 10 GB (upto 30 days)
No limits Win Linux Mac DBsExchange LotusVMware
Ubuntu One Isle of Man 5 GB 50 GB Win Linux AndroidiOS
SafeSync Japan 500 GB (upto 30 days)
No limits Win Mac iOS Android
F-Secure Finland ndash No limits Win Mac
Daten-Safe Austria ndash No limits Win Linux Mac DBsExchange LotusVMWare
and lets the data within the company which underlines the benefits of cloud computingHybrid clouds combine placing a part of the data into a public cloud and processing theother part of data in an own private cloud An example of a cloud backup system is theAmazon Web Services provisioning platform (AWS) which also includes the AmazonElastic Compute Cloud (Amazon EC2) and consequently follows the service-orientedarchitecture principles The Amazon Web Services platform provides access to a largenumber of different further services like application access virtual machines backupof files databases processing queues online-memory (see an overview in Fig 434 andFig 435) Other popular cloud providers with free storage plans are Google Drive [3]Azure [14] and with a focus on processing the Yahoo Cloud [31]
442 RAIC Storage Service Integration
Cloud storage is often used for backups but also for extended storage capacity andsharing of data between devices and users Up-to-date cloud technologies aimed at
118 4 Cloud Computing Virtualisation Storage and Networking
Fig 434 Structure and components of Amazon Web Services
Compute amp NetworkingDirect ConnectDedicated Network Connection to AWS
Deployment amp Management
EC2Virtual Servers in the Cloud
CloudFormationTemplated AWS Resource Creation
CloudWatchResource amp Application Monitoring
Elastic BeanstalkAWS Application Container
IAMSecure AWS Access Control
CloudSearchManaged Search Service
SESEmail Sending Service
SNSPush Notification ServiceSQSMessage Queue Service
SWFWork flow Service for CoordinatingApplication Components
App Services
Elastic MapReduceManaged Hadoop Framework
Route 53Scalable Domain Name System
VPCIsolated Cloud Resources
CloudFrontGlobal Content Delivery Network
GlacierArchive Storage in the Cloud
S3Scalable Storage in the Cloud
Storage GatewayIntegrates on-premises IT environmentswith Cloud storage
Storage amp Content Delivery
Fig 435 Screenshot of the main panel of Amazon Web Services
backup and restore routines of critical enterprise or authority data are discussed in [23]A scheduled comparative analysis of existing complex solutions and standalone tools hasbeen done and represents the advantages of combined (private + public) clouds regarding
44 Backup Services within Clouds as Advanced Cloud BackupTechnology 119
to traditional data-center backups and some known cloud backup solutions In orderto achieve full convenience and elasticity clients require an intelligent combination ofexternally maintained public storage clouds with use of efficient cryptographic methodsand stripesparity dispersal functionality for authenticated transparently encrypted low-overhead and reliable data access This approach has become popular with the nameRAIC ndash Redundant Arrays of Independent Clouds in analogy to RAID One RAIC real-isation is the deployment of the hybrid clouds as a combination of private and publicclouds in certain topologies The combined hybrid clouds with additional cryptographicprotection functionality and management layer (so called ldquocloud storage controllerrdquo) atthe client side is often an appropriate solution Taken to the extreme such setups caninclude peripheral devices such as USB sticks for a four-eye principle in access control Akey point of a hybrid cloud backup concept under the given circumstances is the flexibleconfiguration of all data encoding and decoding steps For increased confidentiality datais transparently encrypted with a symmetric key using for instance the AES cipherFor increased availability data is replicated n times or erasure-coded and subsequentlydispersed The choice and order of data coding and dispersion steps belong to the mainfunctions of an integrating storage service controller [9 25 30]
Many RAIC characteristics can be explained with corresponding RAID methods andliterature In local backup setups the most popular systems are the RAID numbered as0 1 and 5 correspondingly with two or four disks of which zero or one are redundant
The functionality of RAIDs is based on stripes and parity dispersal routines [27] InFig 436 for a RAID5 a representation is depicted The partition in the usual disks array isgiven with different colours firstly for the data (the so called ldquostripe setrdquo eg A1 or C3)and then the distribution of the parity sums (ldquoparity setrdquo eg BP or DQ) through the fivedisks Disk 0 Disk 4 In the given case the common available volume V for the databackup will be calculated with the formula (cp Fig 436c)
V D n 1Vmin (41)
Fig 436 The most used systems RAID 0 1 4 5 6 (RAID) Redundant Array of IndependentDisks (HDD) Hard Disk Drives (up to five disks disk 0 disk 4)
120 4 Cloud Computing Virtualisation Storage and Networking
Whereas n is the number of used HDDs and Vmin the minimal available HDD volume inthe array The redundancy is self-evident preconditioned via the parity set
Example 47 Let us here consider the example with four arrays each of a capacity of500 GByte for RAID5 to find out about the RAID efficiency
V D 4 1 500 GByte
D 1500 GByte(42)
This results in 1500 GB pure for data backup as well as 500 GB for the parity control (seeFig 436c) Therefore a next constructive idea is the deployment of redundant cloud arrays(stripe and parity based dispersion) There are naturally a lot of further RAID conceptsoptimised for minimum access time minimum failure probability maximum volumesminimum costs
Practically these multiple RAID concepts can be continued and mapped to RAICsThere are already numerous subconcepts of RAICs or Redundant Arrays of IndependentClouds The possible variations to the concept are also Redundant Array of IndependentNetworked Storages (RAINS) as well as Random Array of Independent Data Centers(RAIDC) or Redundant Array of Optimal Clouds an extension to RAIC which emphasisesan enforcement of user requirements on the selection and maintenance of storage servicearrays (RAOC) The software architecture suitable for the realisation of RAIC is depictedin Fig 437 The predominant client-side software for RAICs consists of the followingthree layers with the related functionality (1) integration layer (with logical partitionand interface to the backup application) (2) pre-processing layer (with stripes and paritydispersal routine encryption and other modifications) (3) transport layer (with blocktransfer operations) The clients obtain the possibility of the reliable and efficient access toan array of HDD storage media with added organisational and spatial independence Thissoftware considers the state-of-the-art The advanced software architecture realises a newlayered RAIC concept and includes the following already known components but with theextended functionality Firstly the advanced integration layer (1) includes multiple net-work file system protocols like NFS CIFSSMB WebDAV or alternatively a local virtualfile system interface or a Web Services interface Additionally CVSSVNGit (versioncontrol subsystems) and synchronisation overlays are integrated On the other hand anadvanced pre-processing layer (2) consists of necessary codecs aimed to classification ofdocument types and its efficient coding (text files MPEG PDF) Then the policies on thedata storage subjects and paths are included here as well as the routines for stripes andparity dispersion authentication with MDRSAPKI and encryption with AESRSAPKIFinally the advanced transport layer (3) integrates the parallel and block-wise streamingcaching and local persistence procedures as well as includes the adapters for multipleprovider APIs The multi-modal cloud clients (desktops tablets and smartphones) enjoy
44 Backup Services within Clouds as Advanced Cloud BackupTechnology 121
Fig 437 Software architecture of a RAIC
Fig 438 RAID Double Parity structure
with the reliable and efficient resource access to the set of the hybrid (private-public) cloudstorage media namely to the RAIC
RAID DP (Double Parity) is a block-level RAID system with double striping of parityinformation on separated HDDs based on both RAID4 and RAID6 structures The secondparity Q (see Fig 438) can be computed with the same formula as the first parity P butwith other data stripes
122 4 Cloud Computing Virtualisation Storage and Networking
The first parity is horizontal the calculated second parity Q diagonal see formula 43
P1 D XORA1 B1 C1
P2 D XORA2 B2 C2
P3 D XORA3 B3 C3
Q1 D XORP1 A2 B3 0
Q2 D XORP2 A3 0 C1
Q3 D XORP3 0 B1 C2
Q4 D XOR0 A1 B2 C3
(43)
Since in a RAID DP any two disk failures can be compensated the availability of sucha system is increased compared to a single-parity solution The recommended RAID-DPsets consist usually of 14 + 2 HDDs The restoring via RAID DP is relatively simple Thefurther advantages of RAID DP are the simplicity of XOR-Operation for parity computingand possibility to conversion to RAID 4 via switching-off of the Q-stripes Deployment ofoptimised RAID DP offers the advantages as follows
n 5netto
brutto
n 2
nfailuresecurity D 2 (44)
in comparison to well-known RAIC5 (cp Fig 436c)All services offered over the Internet are interacted with according to certain usage
lifecycle phases Storage services are no exception they also adhere to a lifecycleFigure 439 presents the relevant phases and introduces suitable client-side integrationhandlers for each phase The first three phases (discovery and selection contracting andconfiguration) can be subsumed under the term matchmaking These phases typicallyapply once per user-service relationship The fourth phase usage is executed more thanonce and depends on the preceding phases The presented service integration concept is ageneral one For mobile clients bound to storage services in the cloud its interpretation isas follows During the service discovery a dialogue on the device screen guides the userto the right storage services for any given situation By using automation and autonomiccomputing concepts the dialogue can be kept simple or even not be shown at all at theexpense of honouring custom user preferences Then more client-side agents performthe necessary configuration of the services including account creation and registrationwithin the storage controller Finally a scheduler within the storage controller ordersthe timely transmission of data to and from the device Agent frameworks to handle thesign-up to services already exist for example OSST the Online Service Sign-up ToolThe frameworks assume access to a well-maintained service registry which not only
44 Backup Services within Clouds as Advanced Cloud BackupTechnology 123
Fig 439 Live cycle of services
contains information about the services but also links to service-specific agent extensionsHowever the frameworks need to be implicitly parameterised according to the specificneeds of mobile users and with appropriate information already present on the mobiledevice including identities (Fig 439)
In summary the presented background information demonstrates the feasibility ofintegrating storage services on mobile devices in a partially automated process Thenext section will give detailed insight into appropriate choices of methods and theirparameterisation
Hybrid cloud backup concept Figure 440 shows how to transparently encrypt data tobe backed up in a hybrid cloud environment Both a private cloud operated in a user-controlled data centre or across the userrsquos personal devices and a public cloud offered bya commercial or institutional entity can be flexibly combined this way without worryingabout the loss or leak of data
The notion of transparent encryption for cloud backup encompasses the followingfeatures efficient cryptography methods such as AES RSA MDMAC X509Kerberospublic key certificates PKI deployment document classification and demarcation anal-ysis of structured unstructured data and context information user authentication andrespective keys granting
An example of implementation At this point an advanced example of an implemen-tation for the RAIC and RAOC concepts can be mentioned Its origins were in the
124 4 Cloud Computing Virtualisation Storage and Networking
Fig 440 (MD) Message Digest (MAC) Message Identification Code (AES) Advanced Encryp-tion Standard (RSA) Rivest Shamir and Adleman Encryption (PKI) Public Key Infrastructure(X509 Kerberos) Cloud backup and transparent encryption
FlexCloud young investigator group at Technische Universitaumlt Dresden in Germany whichran from 2010 to 2013 The goals of the group were oriented towards a user-controllableand secure cloud life cycle The concrete measures were avoiding uninformed cloudprovider selections through formal descriptions of resource data and software propertiesavoiding the cloud provider lock-in effect through multi-cloud scenarios and migrationpaths towards inter-connected personal clouds under the control of the user which canbe federated into a powerful network of clouds finally means to exert the control withan appropriate management user interface representing a personal cloud cockpit Thisstrategic thinking has influenced the design and development of the file storage solutionNubisave (from Latin ldquoNubesrdquo meaning ldquoCloudrdquo) As project result with the highestpractical value it has subsequently been advanced in the Cloud Storage Lab and is stilloffered for download on this website [34 35]
Nubisave sets up an aggregated view across multiple cloud storage providers andenables higher-level storage tasks such as policy-enforcing data gateways adaptivesynchronisation between devices backup and collaborative sharing Nubisave exportsa virtual file system through the Linux interface File System in Userspace (FUSE)which can be used as an underlay target media of backup software All write accessesreceived by Nubisave are multiplexed onto the configured cloud storage providersand all read accesses reassemble the data Encryption and versioning can entirely beperformed on the client side In case of failures affected storage providers can bereplaced by others and a replication of data from the remaining ones takes placeautomatically Nubisave is available as open source software which has been demonstratedand discussed at both commercial events (trade shows) and academic events (conferencesmeetings)
45 RAIC Integration for Network Storages on Mobile Devices 125
45 RAIC Integration for Network Storages on Mobile Devices
Motivation Systems to combine multiple network and online storage targets withimplied redundancy security and fault tolerance so-called RAICs have recently seenrenewed discussion due to the growing popularity of convenient cloud storage serviceofferings For mobile device access to RAICs less research results are available Aldquosmartphone for the futurerdquo with pervasive storage availability should be intelligentlyand autonomically connected to the cloud Such a constellation allows access withoutgreat expenses to multiple applications data and further resources One necessity is thatthe requirements of the users (security privacy safety pricing and vendor selection) aswell as the functional user objectives are rewarded in the best way In addition valuablebattery capacities need to be saved by selecting appropriate algorithms and parametersand by placing parts of the RAIC integration into the infrastructure On the functionalside for distributed data storage specific resource services with versatile features such asextended storage capacity backup synchronisation and collaborative sharing of data needto be supported The result is a mobile energy-efficient and autonomic RAIC integrationapplication In other words a storage controller on a smartphone
The term Smartphone Bloodbath has been descriptively in use in mobile phone industryreports for the race to more features and lower prices at high frequency for aboutthree years Essentially a phone is technically valued by its hardware functionality andquality its software and services ecosystem and its connectivity Most smartphones offersophisticated software application distribution whereas the innovation in terms of datamanagement is relatively slow The separation between private and business activitiesreflects to some extent on data management and yet most users would need a much morepowerful data and storage feature set One idea for a user-friendly ldquosmartphone for thefuturerdquo is to bind it to online storage services through a pervasive cloud of user-controlledaccounts at registered providers The online storage area allocation would grow and shrinkon demand This binding is similar to how clouds and resource-constrained cyber-physicalsystems and robots are already connected to each other to offload tasks from the devicesinto the network infrastructure One difference between phones and robots is the self-determined nature of user actions When a user records a movie or downloads files thephonersquos media size restrictions will be defused and additional functionality includingonline access to all private data becomes possible although the user may decide to overridethe use of the online storage The binding to multiple services at once requires intelligentclient-side integration techniques with phase-of-lifecycle knowledge which additionallymatch the service properties against user requirements For secure and reliable datastorage the RAIC concept has been proposed as integration technique and successfullyimplemented for desktop computers and enterprise storage integrators [29] However froma security and convenience perspective on mobile devices the RAIC assembly and thedistribution of the data to the attached providers needs to happen directly on the deviceitself in most cases which contradicts a conservation of battery power It is therefore
126 4 Cloud Computing Virtualisation Storage and Networking
important to integrate network storage services on mobile devices in a systematic way forpredictable storage characteristics even under changing networking and device conditions
In the next sections the basic concepts behind network and cloud storage RAICsand their applications including hybrid backup clouds are presented The phases of theusage lifecycle of services in general and storage services in particular are examinedin detail to derive a suitable integration design Tradeoffs between user-friendly fullautomation and control-preserving semi-automatic or guided integration are discussed inthis context Intelligent RAIC use in the mobile field further implies certain decisionson which algorithms parameters and placement strategies to use in order to preservethe battery and gracefully adapt to imperfect networking conditions The next part istherefore outlining specialised data coding techniques including encryption splittingerasure codes and all-or-nothing transformations Again tradeoffs need to be understoodcorrectly to achieve high-performance integration with low power consumption Thepeculiarities of mobile access to RAICs are shown using elaborated software architec-ture on a selected smartphone platform Finally a summary of the findings and anoutlook on further ideas to improve the connections of smartphones into the cloud isgiven
451 Efficient Access to Storage Services from Mobile Devices
Depending on the use cases the weight of comparison parameters to distinguish themost suitable RAIC integration method differs For many client systems security playsa major role and motivates distributed data storage with comparatively more storageoverhead in return for higher security As a generalisation thereof subjectively optimalparameters including storage and retrieval times and service costs can be considered andweighted by clients at configuration time yielding RAOCs [33] For mobile devices twoparameters become dominant The energy efficiency of the integration and the usabilityunder imperfect networking conditions Both have so far not been subject to analysisfor the research on RAICs but are crucial for the further acceptance of such techniquesEnergy efficiency can be broken down into the (negligible) setup service selection signupand configurationreconfiguration processes which typically donrsquot happen more than onceper device power-on session and the service usage processes for storing and retrievingdata Measuring the energy efficiency of algorithms requires specialised equipment Theelectrical power consumption is not linear to the performance but grows along with ithence a performance comparison assuming equal processor load can be used for a firstestimation The power consumption analysis in this example are made using the HAECndash Highly Adaptive Energy-Efficient Computing measurement infrastructure as shown inthe photo below (Fig 441)
Performance characteristics of RAIC integration techniques based on [32 35] aresummarised in Table 49
45 RAIC Integration for Network Storages on Mobile Devices 127
Fig 441 HAEC laboratory measurement equipment (own photo)
Table 49 Qualitative comparison of performance characteristics for versatile RAIC integrationtechniques
Technique Read performance Write performance
RS erasure code 0 redundancy XOR 100 100
RS erasure code 0 redundancy SIMD 270 ndash1200 270 ndash1200
RS erasure code 50 redundancy n = 3 100 67
AONT-RS n = 3 33 33
Imperfect networking usability mandates an intelligent use of caching and schedulingso that slow or broken links will show no or little effect on the user of a RAIC Thistypically differs per implementation However already on the algorithmic level someerasure codes have been more optimised for storage retrieval and repair than othersResearchers have identified suitable algorithms through experiments [26] Based on theseobservations we can assume that the use of processor-specific erasure codes is beneficialfor mobile devices Both the devicersquos energy efficiency and the imperfect networkingusability can be tremendously improved by placing the RAIC integration onto a trustedlocal network proxy So-called storage integrators can serve multiple users and enforcegroup policies On the other hand they have drawbacks concerning the trust mobilityand overall energy efficiency given that such additional devices will remain idle forlong durations Figure 442 shows both possible integration approaches in a comparisonarchitecture scheme
128 4 Cloud Computing Virtualisation Storage and Networking
Fig 442 Variants for efficient placement of RAIC integrator between the clouds
452 A New Must-Have App RAIC Integrator for Smartphones
While our results are generally applicable to all mobile devices including tablets andnotebooks our realisation scenario focuses on mobile phones due to their increasingpopularity as ldquoswiss army knivesrdquo for computing tasks Today such phones ship withinternal storage media (ROM non-volatile flash memory SD cards) and otherwise rely onmanual storage service integration beyond the sometimes preconfigured vendor-specificservices Increasing amounts of data produced by mobile phone sensors and applicationspush the idea of a ldquosmartphone for the futurerdquo with ubiquitous access to elastic storage inthe cloud Such a smartphone requires inter alia an operating-system integrated library fortransparent RAIC integration across all applications which need extended storage capacityoffsite backups and other uses of storage Essential parts of the integrator are (1) a databasewith information about available services including their functional and non-functionalproperties and protocols for accessing them (2) protocol-specific access modules (3) adispersion module which splits the data according to the user-defined parameters whileconsidering energy efficiency and imperfect networking conditions and (4) autonomicsupport functions for service sign-up and repair in case of failures The binding of a mobilephone to a RAIC-DP configuration through an integrator is depicted in Fig 443 TheP-stripe is stored in the private cloud client while the Q-stripe is delegated to the publicclouds ie to the provider Arbitrary RAIC and dispersion configurations are possiblealthough certain key configurations will be preferred by mobile users RAIC-DP for highestsafety AONT for highest (information-theoretic) security and JBOCRAIC0 for the leastamount of overhead A configuration wizard would have to present these choices to theusers in a meaningful way
Suitable software architecture for the realisation of a mobile RAIC over both local andcloud storage resources is depicted via Fig 444 following the design proposed for genericcloud storage controllers The predominant client-side software for RAICs consists of thefollowing three layers with the related functionality
45 RAIC Integration for Network Storages on Mobile Devices 129
Fig 443 RAIC-DP A network storage model
Fig 444 Offered software architecture to realisation of a RAIC (HDD) Hard Disk Drive orother local drives including SD media (RAIC) Redundant Arrays of Independent Clouds (CVS)Concurrent Versioning System
130 4 Cloud Computing Virtualisation Storage and Networking
1 Integration layer logical partition and interface to the backup application2 Pre-processing layer stripesparity dispersal routine encryption and other modifica-
tions3 Transport layer block transfer
The clients obtain the possibility of reliable and efficient access to an array of virtualisedstorage media offered as a service or as local complementary media with addedorganisational and spatial independence This software considers the state-of-the-art Theoffered software layered architecture realises a RAIC concept and includes the followingalready known components with the extended functionality
1 Advanced integration layer A local virtual file system interface available to allapplications Depending on the operating system there may be additional specificinterfaces for instance the registration as content provider on Android or the exportas RESTful web service through RestFS
2 Advanced pre-processing layer Codecs classification of document types andcoding (text files MPEG PDF) Policies on the data storage subjects and pathsStripesparity dispersion routines Authentication with MDRSAPKI Encryption withAESRSAPKI
3 Advanced transport layer Parallel and block-wise streaming Caching and localpersistence Adapters for multiple provider APIs
The proposed system can be implemented with existing academic and open sourcesoftware Nubisave [33] is a cloud storage controller which performs the functionalityof the upper layer as a Linux user-space file system (FUSE) module with 1 file inputand n fragment outputs Through the Nubisave configuration GUI the remaining twolayers can also be controlled For instance the Nubisave splitter modulersquos first outputcan be connected to an EncFS module for data encryption which is in turn connected to aFuseDAV module for placing the encrypted fragment data on a protected WebDAV folderwhich serves as standard-compliant interface to a cloud storage area
Some mobile phone operating systems run directly on Linux including Maemo and themore recent SailfishOS and FirefoxOS so that Nubisaversquos file system interface is a suitablemeans for data access across all applications For Android and similar systems withrestricted global data access a translator between files and the respective per-applicationcontent API would be required Imperfect network handling is an implementation detailof the transport modules We have previously refined fault-tolerance access to RESTfulservices (including eg WebDAV as HTTP extension) as RAFT-REST concept The JavaResUp library [38] is available to transport module authors as a convenient caching andretransmission handler Beyond the specific transport modules Nubisave also caches databy itself to some extent Hence the combination of a cloud storage controller with energy-efficient parameterisation agent-based service lifecycle handling for semi-automatic
References 131
integration and fault-tolerant service integration under imperfect networking conditionsis possible today and fulfill the requirements of mobile users
The next problems to solve are
bull Analysis of integration options for existing cloud storage services (Cloud-of-Clouds)bull RAIC Cloud backup concept elaboration (stripe and parity based dispersion)bull Development of software RAIC controllers based on web services for management and
cryptographic protection of a RAIC (combined clouds) eg RAIC5 RAIC-DPbull Deployment of proxy servers for easy mediationbull Development and securing the meta-data database for RAIC managementbull Development of easy-to-use conditions a common access scheme for the enterprises
with offering of good performance high security data control for the usersbull Further development of collaboration scenarios file sharing access by external entities
CVS and group working automatic classification of databull Improving performance eg scheduling algorithms cachingprefetching and paralleli-
sation
46 Conclusions
This chapter has given a brief systematic introduction into the challenges of operatingand integrating cloud services related to computing resources computation networkand storage It has covered recent trends including distributed storage facilities for highavailability and confidentiality integration of cloud services into mobile devices with highenergy efficiency and pervasive or ubiquitous access to multiplexed cloud services Forsmartphone makers the results show that especially storage integration is a desirablefeature which leads to outstanding devices with a functionality closer to what highlydemanding users expect
References
1 Amazon Web Services online httpawsamazoncom 20132 Citrix Systems ShareFile online httpwwwcitrixcomproductssharefileoverviewhtml
20133 Google Drive online httpsdrivegooglecom 20134 Ordinary backup technologies online httpwwwtecchanneldestoragebackup 2015 in
German5 Security Compendium online httpwwwsecurity-insiderde 2015 in German6 Ukrainian legislation regarding to data security online httpzakonradagovua 20157 VMware vSphere API for Storage Awareness online httpwwwvmwarecom 20138 C Baun M Kunze J Nimis and S Tai Cloud computing ndash Web-based dynamic IT-Services
Springer-Verlag 2010 in German
132 4 Cloud Computing Virtualisation Storage and Networking
9 G R Blakley Safeguarding cryptographic keys In AFIPS Conference Proceedings volume 48p 313ndash317 1979 National Computer Conference (NCC)
10 D Decasper A Samuels and J Stone RAIC ndash Redundant Array of Independent Clouds patentUSA Reg No 12860 810 Publishing No US 20120047339 A1 2012
11 S Gross J Spillner and A Schill FlexCloudTUD Project Dresden University of TechnologyTUD online httpwwwflexcloudeu 2013
12 Sheikh M Habib and S Hauke and S Ries and Max Muumlhlhaumluser Trust as a Facilitator in CloudComputing A Survey Journal of Cloud Computing Advances Systems and Applications June2012
13 H Kim N Agrawal and C Ungureanu Revisiting Storage for Smartphones ACM Transactionson Storage 8(4) November 2012
14 H Kommalapati Windows Azure Platform for Enterprises online httpmsdnmicrosoftcomen-usmagazineee309870aspx 2013
15 Thomas A Limoncelli OpenFlow A Radical New Idea in Networking Communications of theACM 55(8)42ndash47 2012
16 A Luntovskyy and D Guumltter A Concept for a Modern Virtual Telecommunication EngineeringOffice International Research Journal of Telecommunication Sciences 3(1)15ndash21 2012
17 A Luntovskyy and M Klymash The service-oriented Internet In Proceedings of IEEE 11thTCSET 2012 Conference on Modern Problems of Radio Engineering Telecommunications andComputer Science 2012 Lviv ndash Slavsk Ukraine
18 A Luntovskyy M Klymash and A Semenko Distributed services for telecommunicationnetworks Ubiquitous computing and cloud technologies Lvivska Politechnika Lviv Ukraine2012 368 p Monograph in Ukrainian
19 A O Luntovskyy Programming Technologies of Distributed Applications DUIKT StateUniversity of Telecommunications Kyiv 2010 474p in Ukrainian
20 A O Luntovskyy M V Zakharchenko and A I Semenko Multiservice Mobile PlatformsDUIKT State University of Telecommunications Kyiv 2015 216p in Ukrainian
21 Andriy Luntovskyy Dietbert Guumltter and Igor Melnyk Planung und Optimierung von Rechner-netzen Methoden Modelle Tools fuumlr Entwurf Diagnose und Management im Lebenszyklus vondrahtgebundenen und drahtlosen Rechnernetzen SpringerVieweg + Teubner Verlag Wiesbaden2011 411 p in German
22 Andriy Luntovskyy and M Klymash Data Security in Distributed Systems LvivskaPolitechnika Lviv Ukraine 2014 464 p Monograph in Ukrainian
23 Andriy Luntovskyy Volodymyr Vasyutynskyy and Josef Spillner RAICs as Advanced CloudBackup Technology in Telecommunication Networks International Research Journal ofTelecommunication Sciences 3(2)30ndash38 December 2012
24 P Mell and T Grance The NIST definition of cloud computing whitepaper NIST SpecialPublication 800ndash145 September 2011
25 J S Plank S Simmerman and C D Schuman Jerasure A Library in CC++ FacilitatingErasure Coding for Storage Applications ndash Version 12 Technical Report CS-08-627 Universityof Tennessee 2008
26 J S Plank K M Greenan and E L Miller Screaming Fast Galois Field Arithmentic UsingIntel SIMD Instructions In Usenix FAST February 2013
27 M O Rabin Efficient Dispersal of Information for Security Load Balancing and FaultTolerance Journal of the ACM 36(2)335ndash348 1989
28 Johannes Schad Stephan Zepezauer and Josef Spillner Personal Cloud Management Cockpitwith Social or Market-Driven Asset Exchange In Networked Systems Conference (NetSysKiVS)ndash Communication Software Award Demo March 2013 Stuttgart Germany (Vorfuumlhrung)
References 133
29 Ronny Seiger Stephan Groszlig and Alexander Schill SecCSIE A Secure Cloud Storage Integratorfor Enterprises In International Workshop on Clouds for Enterprises (C4E) p 252ndash255September 2011 Luxembourg Luxembourg
30 A Shamir How to Share a Secret Communications of the ACM 22(11)612ndash613 197931 Shelton Shugar Cloud Computing at Yahoo online httpopencirrusorg 201332 C A N Soules G R Goodson J D Strunk and G R Ganger Metadata efficiency in
versioning file systems In Proceedings of the Third USENIX Conference on File and StorageTechnologies April 2003 San Francisco California USA
33 Josef Spillner Gerd Bombach Steffen Matthischke Johannes Muumlller Rico Tzschichholz andAlexander Schill Information Dispersion over Redundant Arrays of Optimal Cloud Storage forDesktop Users In 4th IEEEACM International Conference on Utility and Cloud Computing(UCC) p 1ndash8 December 2011 Melbourne Australia
34 Josef Spillner and Johannes Muumlller PICav Precise Iterative and Complement-based CloudStorage Availability Calculation Scheme In 7th IEEEACM International Conference on Utilityand Cloud Computing (UCC) p 443ndash450 December 2014 London UK
35 Josef Spillner Johannes Muumlller and Alexander Schill Creating Optimal Cloud Storage SystemsFuture Generation Computer Systems 29(4)1062ndash1072 June 2013 DOI httpdxdoiorg101016jfuture201206004
36 Josef Spillner Christian Piechnick Claas Wilke Uwe Aszligmann and Alexander SchillAutonomous Participation in Cloud Services In 2nd International Workshop on IntelligentTechniques and Architectures for Autonomic Clouds (ITAAC) p 289ndash294 November 2012Chicago Illinois USA
37 Josef Spillner and Alexander Schill A Versatile and Scalable Everything-as-a-Service Registryand Discovery In 3rd International Conference on Cloud Computing and Services Science(CLOSER) p 175ndash183 May 2013 Aachen Germany
38 Josef Spillner Anna Utlik Thomas Springer and Alexander Schill RAFT-REST ndash A Client-side Framework for Reliable Adaptive and Fault-Tolerant RESTful Service Consumption In2nd European Conference on Service-Oriented and Cloud Computing (ESOCC) volume 8135of LNCS p 104ndash118 September 2013 Maacutelaga Spain
5Smart Grid Internet of Things and Fog Computing
Keywords
Integration of networks for telecommunications and energy supply bull New servicearchitectures bull Demarcation of grid vs smart grid bull Power Line Communication(PLC) bull Green computing bull Energy-efficient communication (Bluetooth bull Zig-Bee bull EnOcean bull 6LoWPAN) bull Demarcation of Internet of Things (IoT) vsInternet of Services (IoS) bull Fog computing bull Distributed computing bull Mini-PC bull On-board -controllers (Raspberry Pi bull Arduino) bull Computer-Aided Design(CAD) bull Automation networks bull Smart home bull Smart factory bull Industry 40
In the previous chapters we have highlighted the evolution of computing environmentsfrom single systems to parallel architectures clusters grids service-oriented systems andclouds This line of evolution is a purely digital one without considering the form factorof computing From the physical perspective there is another line of evolution whichputs the form factor and communication channels into the centre Starting with mini-PCsand embedded computers nowadays distributed computing can be performed in wearablecomputers and body-area networks tiny nodes organised as fogs or smart dust connectedto the Internet of Things and in the ldquoSmart Gridrdquo using various protocols This chaptertherefore outlines physical computing paradigms and compares the computing storageand communication capabilities
The first part of the chapter examines some typical scenarios for ldquoSmart Gridrdquodeployment as a technology for integration of electric power supply networks and telecom-munication networks and gives a comparison of corresponding services architectures andmulti-level models The smart grid enabling technologies are discussed Furthermoresome case studies on smart grid are presented
copy Springer Fachmedien Wiesbaden GmbH 2017A Luntovskyy J Spillner Architectural Transformations in Network Services andDistributed Systems DOI 101007978-3-658-14842-3_5
135
136 5 Smart Grid Internet of Things and Fog Computing
The second part of this chapter is dedicated to the up-to-date development of the IoTand of ldquoFog Computingrdquo based on the achievements in Wireless Personal Area Network(WPAN) The IoTfog computing enabling technologies are discussed Case studiesbased on use of on-board -controllers such as Raspberry Pi are examined
51 Smart Grid as Integration Technology for the Networks ofEnergy Supply and Telecommunication
Green computing Green IT is an important predecessor trend and part of smart griddevelopment because of the significant efforts on waste heat and energy recycling(Fig 51) Additionally to the known Power Usage Effectiveness (PUE) the EnergyReuse Efficiency (ERE) value has to be used Let us compare one to each other
Power usage effectiveness (PUE)
bull PUE D TotalFacilityEnergy=ITEquipmentEnergy D Ptotal=PIT
bull PUE gt 10 ideal value 101
bull compare to K D PIT=Ptotal D 1=PUE0 lt K lt 1
Energy reuse efficiency (ERE)
bull ERE D TotalEnergyConsumptionofaDataCentre RecyclingEnergy
=ITEquipmentEnergy D P0
total Precycling=PIT
bull 0 lt ERE lt PUE lt 150
Fig 51 Green IT symbolic representation (Source magatechnologyit)
51 Smart Grid as Integration Technology for the Networks of Energy 137
Fig 52 The construction of the efficient energy consumption and recycling within of a buildingwith a lsquogreenrsquo IT data centre (According to [39])
Example 51 For one particular provider of green IT services [39] the following valueshave been computed PUE D 102=105 and ERE D 062=068 The high efficiency isbased mainly on the water cooling and the renunciation of all refrigeration systems Theconstruction of the efficient energy consumption and recycling within of a building withgreen IT data centre (according to [39]) is given in Fig 52 The construction encompassesthe following components (1) servers (2) tank and warm water preparation (3) airwaterwarm pump (4) air supply (5) ventilation (6) air supply to the servers (7) ventilationfrom the servers (8) air supply warm pump (9) ventilation from warm pump (10)heating forerun (11) heating countercurrent (12) warm water (13) fresh water stream(14) heating system (15) warm water cone filters and (16) optional photovoltaic
The computation of the energy efficiency is given in Fig 53 According to thiscomputation the following PUE values are resulting
1 PUE without the warm pump PUE D 254ŒMW=a=250 6ŒMW=a D 101 ie254ŒMW=a D 262 8ŒMW=a (for warm pump)
2 PUE with the warm pump PUE D 105 ie 262ŒMW=a=250 6ŒMW=a
138 5 Smart Grid Internet of Things and Fog Computing
Fig 53 The computation of the energy efficiency (Source [39])
Smart grid definition The high-tech twenty-first century is in particular also thecentury of ldquosmall power supply systemsrdquo due to the use of advanced information andcommunication technologies in energy networks Creation of combined systems calledldquoSmart Gridrdquo opens great prospects for the development of both of these industries (energyand IT) and is intended to provide a synergistic effect This section examines existingmodels of smart grid suitable basic networking technologies as well as typical usagescenarios for integrated intelligent networks
Smart grid is a technological direction for the integration of electric power supplyand telecommunication networks in order to increase the energy efficiency of bothtypes of networks reduction of CO2 emission under the Kyoto Protocol consideringdecentralisation of existing architectures for an integrated network (ie one of the mainprinciples of Internet construction) and improving of its efficiency (efficient switchingrouting) under use of alternative and renewable energy sources (like wind solar Electro-Magnetic (EM)-smog) combined with use of hybrid hydrocarbon-electric vehicles(Plug-in (Hybrid) Electric Vehicles (PEV)) with optimisation of network managementtechniques and billing services (Smart Metering) within the conventional power supplynetworks as well as increasing its safety security and Quality of Service (QoS) in suchintegrated networks for power supply and telecommunication [23 35]
The conceptual terms laquogridraquo and laquosmart gridraquo should not be confused The (intelli-gent) grid network solutions are used for time-consuming computing tasks (simulationplanning forecasting etc) based on (virtual) server clusters or supernodes with use ofconventional protocols of the Internet Nowadays grids are a weighting part of innovativecloud computing technology (for instance by IaaS Infrastructure as a Service) [30] whenthe (mobile) client access to computing power is very easy The most important task whichhas already become a ldquoclassicrdquo of the grid technology is a rational and decentralised
51 Smart Grid as Integration Technology for the Networks of Energy 139
redistribution of computational workload between participating (virtual) servers clustersor supernodes in the computing life-cycle of time-consuming engineering scientific oreconomical tasks
Therefore the concepts of laquogridraquo and laquosmart gridraquo are co-related areas of researchBut the energy efficiency is not a direct scientific and technical challenge for purelycomputational grid technology [9] Heat and redundant energy occurs here only asby-product and even a harmful product (ldquoheat waste productsrdquo of modern networktechnology)
Active deployment of the environmentally friendly and thus laquogreenraquo smart gridtechnology goes on today in many developed countries for example Australia EuropeanUnion in particular Germany and Austria USA Canada Peoplersquos Republic of China andSouth Korea which would like to provide and reinforce their own energy independencefor the future Several leading research groups in universities carry out the correspondingresearch subjects on the mentioned area and already possess certain ldquoknow-howrdquo oftenin combination with innovative companies An example is the Kiwigrid Smart GridManagement Platform engineered in cooperation with Technische Universitaumlt Dresdenin Germany (TUD) [10] On this platform energy applications are offered through amarketplace and linked to data processing and analytics services A smart meter gatewayconnects devices and allows for an optimisation of power consumption
The slogan of the coordinated actions might be for all stakeholders as follows ldquoFromthe Internet of Data and Web Services to the Internet of Energy Servicesrdquo Nowadays thereare numerous international organisations and well-known companies that are developingthe technology and corresponding devices for smart grid Among them are IEEECENELEC Cisco Deutsche Telekom Siemens etc [2 4 9 16 21]
The existing basis for local-area solutions of smart grid is built on the followingwell-known network technologies Powerline Homeplug Worldwide Interoperability forMicrowave Access (WiMAX) PoE (Power over Ethernet) KNX LON (Local OperatingNetwork) Wireless Sensor Networks (WSN) (ZigBee EnOcean) etc [2842] But thereis also a necessity to develop integrative solutions for network decentralisation (one ofthe main principles of Internet construction) to improve its efficiency to facilitate use ofalternative and renewable energy sources (like wind solar EM-smog) and to stimulate thedevelopment of efficient energy storages (batteries peculiar energy depot) aimed to storeredundant or excess (electric) energy
To reach this goal we first need to formulate a list of scientific and technicaldevelopment challenges for an integrated network (smart grid) on the existing basis ofstandard network architectures then requirements for such networks and then to developits own basic models How will everything work together Consider the following twoscenarios
Example 52 What will be a middle-class network connection for a Small and MediumEnterprise (SME) in 2020 Only one cable or wireless link will provide the utilityservices such as electricity telephony Internet digital high-definition television and cloud
140 5 Smart Grid Internet of Things and Fog Computing
services Room heating will be realised via derivation and recycling of redundant energyfrom multiple (virtual) servers The wired and wireless automation of local-area as wellas piconets like Local Operating Network (LON) KNX Home and Building ControlStandard (KNX) ZigBee EnOcean will be used to serve and control the in-doorclimate Management of such integrated networks can be performed through EthernetLocal Area Network (LAN)Wireless Local Area Network (WLAN) links as wellas convenient protocols like Internet Protocol (IP) Internet Control Message Protocol(ICMP) Simple Network Management Protocol (SNMP) The program supportconfiguration and tuning of the intelligent network is realised with the use of mobiledevices (smartphones and tablets) mobile applications and through offered web servicesrunning in a cloud environment
Example 53 The scenario depicts a vision similar to one particularly involved Germancompany Siemens but is applicable to other companies with a similar portfolio Accord-ing to Fig 54 in the future smart grid is designed to connect four major components [16]which operate both as consumersproducers and electric energy storages The componentsare
Fig 54 AC ndash Alternating Current HVAC ndash Heating Ventilating and Air Conditioning PEV Smartgrid technology highlights inspired by Siemens
51 Smart Grid as Integration Technology for the Networks of Energy 141
1 Intelligent buildings2 Electricity plants3 Electromobility4 Smart metering
Intelligent buildings also called intelligent homes for residential buildings are equippedwith solar panels and local-area networks for climate automation like Field Bus and WSNThese are connected to power plants enterprises for the generation of (electric) energyalso called AC plants based on conventional or alternative and renewable energy sourceslike wind solar and EM-smog Electric mobility based on hydrocarbon-electric hybridvehicles (PEV) that accumulate power and can afterwards ldquouploadrdquo it to the network leadto a strong electromobility Intelligent counters and meters for smart metering automate thecharging and billing processes They carry out the monitoring and network managementaimed at low-energy consumption on the basis of improved tariff models with respect tothe workload parameters and traffic both in analogy to packet-switched networks
The considered components 1ndash4 may both use and release the excess (electro-)energyand stored redundant currents in the network This leads to synergy effects betweenthe different consumers and producers of energy as the timing of the production andconsumption peaks differ widely Furthermore information technology helps to controlthe timing by being able to shift the peaks according to schedules An example is overnightdishwashing which can be programmed to happen at a particularly convenient time basedon electricity supply and cost
Electricity demands and ldquoGreen ITrdquo today Increasing demands of energy and signif-icant rising of ICT prices evoke the necessity of energy use efficiency which has to berealised over the entire IT life cycle ldquodesign ndash operation ndash modification ndash operation ndash rdquoThe ecological protection of the environment CO2 emission discharge economisationof the fossil resources and electricity power plays a very important role nowadays Theenrollment of renewable energy resources is required in operation of facility grids inoperation of IT and networks in disposal of waste energy and in the deployment of smartmeters for the user provider and equipment as well as power plants The correspondingenergy demands per annum by the years 2000 until 2015 are exemplarily shown for theserver and data centres in Germany in Fig 55
Based on the studies in the years 2010ndash2012 of the Borderstep Institute the followingthree tendencies became apparent
1 Since 2008 more and more attention is payed to the ldquoGreen ITrdquo solutions2 Considering as reference the year 2011 we can constitute that the electricity consump-
tion for the server and data centres in 2011 is approximately 14 TWh under the awaiteddemands within the ldquobusiness as usualrdquo In comparison to the ldquoGreen ITrdquo scenario the
142 5 Smart Grid Internet of Things and Fog Computing
16
14
12
0
8
6
4
2
02000
TW
hye
ar
2001
398 TWh
101 TWh
97 TWh
60 TWh
93 TWh
142 TWh
2002 2003
Green IT Business as usual Trend
2004 2005 2006 2007 2008 2009 2010 2011 2012
Borderstep 2010 ndash 2012
2013 2014 2015
Fig 55 Annual tendencies to electricity consumption for server und data centres in Germany
Table 51 The overall annual electricity demands in Germany
Year Electricity demands (gross) Primary (fossil) energy consumption
Overall Renewable energy resources Overall Renewable energy resources
[TWh] [PJ]
1991 5396 32 146 13
2000 5796 66 144 29
2005 6141 102 146 53
2009 5813 163 135 89
2015 600 326 133 125
demands are more than 23 TWh below despite of huge growth of the server and datacentres with significant reducing of the electricity costs of about 12 mia Euros (2011)
3 These partial electricity demands (97 TWh) are corresponding to approximately18 of the overall electricity consumption in Germany To compare to producethe mentioned amount of electricity four middle-dimensioned coal power plants arerequired
The overall annual electricity demands in Germany for some selected years are shown viaTable 51 The representation is based on the sources [1 41]
51 Smart Grid as Integration Technology for the Networks of Energy 143
Fig 56 Forecast for the annual electricity consumption of telecommunications and IT branch inGermany
The simple empirical formula 51 can be taken into account for recalculating ofelectricity volumes With this formula and the given analysis a forecast for the annualelectricity consumption for telecommunications and IT branches until the year 2025 canbe calculated (Fig 56)
1 TWh D 03 PJ (51)
The forecast has shown that the annual electricity consumption of communicationand information businesses in Germany was significantly reduced since 2010 until 2015from 560 TWh down to 478 TWh ie approximately by 15 This important reductiontrend will be continued until the year 2020 and then stabilise at around 452 TWh
462 TWh in 2025 Therefore the positive development of electricity consumption of theseindustries can be distinguished In the given internal structure the cause of this overalldeclining trend becomes clear Successively the electricity demands in households publicand workspace IT areas are reduced In contrast the electricity demands for the dataand computing centres will be increased too based on the increasing data traffic Thisprognosis has foreseen a lot of implemented energy efficiency measures because of greatsocial meaning of ldquoGreen-ITrdquo processes in industrialised countries
How to advance and deepen the success of ldquoGreen ITrdquo in such countries There is thegreat variety of the possible approaches to smart grid implementations as follows
bull videoconferencing instead of business travelbull notes electronically (paperless) instead of on paper
144 5 Smart Grid Internet of Things and Fog Computing
bull reduction of unnecessary printingbull reduction of energy consumption in the use and productionbull sustainable product design and manufacturing durable as possible hardwarebull resource-saving programming (Green Software Engineering)bull reduction of CO2 emissionbull decentralisation of the networkbull QoS increase (efficient switching routing)bull use of alternative and renewable energy sources (wind solar thermal)bull optimisation of measurement and network management (smart metering and energy-
efficient web services)bull increase of network security safety and reliability
511 Services Architectures and Multi-level Models
The integrated architecture of smart grids has to repeat in a certain extent the well-known Open Systems Interconnect (OSI) network architecture (Fig 57) But it mustbe also multi-dimensional ie has to reflect not only the abstraction levels with multipledefined interfaces functions and services but the various types of network technologiesand domains of its use types of consumers and service providers device types accesscontrol techniques schemes to billing and payment for the consumed services
Fig 57 APL ndash Application NWK ndash Network MAC PHY ndash Physical A simplified architecture forsmart grids
51 Smart Grid as Integration Technology for the Networks of Energy 145
Let us consider a selection of the existing multi-layered and multi-dimensional modelsfor smart grid which are oriented towards shared use of telecommunications
1 National Institute of Standards and Technology USA (NIST) Smart Grid ConceptualModel
2 IEEE Smart Grid Model3 A proprietary model of Cisco Smart Grid4 Common architecture of ITGVDE Smart Grid (Germany)5 Next development of model (4) the EU Smart Grid Architecture Model (European)
One of the first models developed in the area the model (1) called NIST Smart GridConceptual Model provides abstraction of properties of the integrated intelligent networkbased on a classic three-level representation encompassing the following levels (1) Powerand Energy (2) Communications (3) IT and Services [11]
The universal model (2) was engineered in IEEE working groups IEEE Smart Gridis a professional organisation for standardisation and co-ordination among the smartgrid stakeholders within IEEE Universality of the mentioned IEEE smart grid modelconsists in the creation and description of a meta-system called smart grid whichextends the rules interfaces and functions for individual intelligent networks to theso-called smart grid domains also based on the following three levels (1) Power andEnergy (2) Communications and finally (3) IT and Services IEEE shifted the focusof consideratioon to the second and third layer as the determining levels for the first layerelectricity distribution in smart grids [9]
The following proprietary model (3) was provided by the company Cisco one of theleading companies in the field of network technologies and products [2] The modeltakes into account the development aspects of integrated (mobile) power transmissionand telecommunications in the context of hardware and software that is produced by thecompany Nowadays Cisco provides design and implementation deployment and supportof infrastructure and services for smart grids as well as numerous communication systemsfor the power supply sub-stations automation networks (Field Area Networks) for powersupply nets provides data security (Cisco Switches Routers Firewalls like ASA-CX) forthe smart grid creates the virtual storage centres for data processing (network storagescloud computing) thus extending those capabilities of Wireless Area Network (WAN)architectures The Cisco Connected Grid Network Management Solutions (NMS) offer theinfrastructure access tools monitoring and management facilities for IP-enabled devicesintegrated into smart grid
Furthermore let us consider the advantages of a common architecture for smartgrid architecture proposed by ITGVDE Existing network technologies can be easilyintegrated into the framework of model (4) The installed services are independent of thebasic network infrastructure (as is the idea of OSI) The common architecture for smartgrids allows adequate modeling of integrated networks of energy and information supplyat different levels of abstraction Model (4) of smart grids can be used recursively or
146 5 Smart Grid Internet of Things and Fog Computing
Levels Smart Power Grid Services
PortalUsers
Smart PowerWeb Services
MarketPlace Portfolio
TechnicalServices
Standardized
Middleware
Proprietary
NodesComm andtransportchannels
VirtualTools
NW
NW
GW ServiceProduction
Tools resourcesAU Automation
MonitoringAC Energy
SupplyNWTelco
Metering
Metering
Metering
AC
AC
AU
AU
Consumers1
2
3
4
Fig 58 GW ndash Gateway AC ndash Alternating Current (energy supply nets) AU ndash Automation (andmanagement) networks SPGWS ndash Smart Power Web Services NW ndash Network Metering ndash controland telemetry Market Place ndash allocation and reselling of services Common 4-layer architecture forsmart grid [18] and the types of energy supply and data supply services (1) consumers (2) servicesand virtualisation (3) info-objects and service communication (4) infrastructurephysical layer
hierarchically to describe the inter-operability between different providers offering theirservices (Fig 58)
bull Communications in particular mobile communicationsbull Electrical energy supplybull Smart metering including intelligent control and telemetrybull Smart power web services
A typical service for smart power distribution would be the brokering of excess productionin households ie micro-plants In such scenarios power is produced by roof-topsolar installations private wind turbines as well as thermal sources Depending on thecompensation of feeding energy into a grid profit for selling it to a nearby user or abenefit from using it for custom purposes such a brokering service guides the producer ofelectricity to a decision
51 Smart Grid as Integration Technology for the Networks of Energy 147
Fig 59 Domains DER ndash Distributed Energy Resources GTD ndash Generation Transmission Distri-bution (production) CP ndash Customer Premise (delivery) Zones Process Field Station OperationEnterprise Market (PFSOEM) EU Smart Grid Model and Architecture [6] (1) business layer (2)function layer (3) information layer (4) communication layer (5) component layer
The presence of the common architecture of smart grids provides nevertheless a widefield for activities and describes the ability of the model to absorb innovations [5 18]
As the development of this well-known and recognised model (4) progresses a morecomplex multi-dimensional European model (5) called EU Smart Grid Architecture(Fig 59) should be considered The model possesses its five component layers as followsBusiness Function Information Communication and Component as well as two furtherdimensions called Domains and Zones [6 22]
Example 54 From the viewpoint of the telecommunications department at DresdenUniversity of Technology [26] ldquo in a green world renewable energy sources are the keyto reduce the CO2 footprint These energy sources are typically non-stationary This factorrequires much more complex control of the grid To enable this the energy distributionnetwork has to become more intelligent due to new services distributed generation ofenergy (virtual power plants) and new safety and security requirements It will finally
148 5 Smart Grid Internet of Things and Fog Computing
Fig 510 LV ndash low voltage MV ndash middle voltage (1) MV part of substations (2) LV part ofsubstations (3) street cabinets (4) substations (MV+LV) (5) interruptions (open meshes) Smartgrid representation as a PLC
be a Smart Gridrdquo Nowadays new demands on reliability and security to the supportcommunication network appear The discussed approach enables close system integrationoptimal distributed power generation via virtual power plants efficient control on theelectricity distribution and deployment of new network services which are becomingmore intelligent simultaneously It has been proven that a particular attention should bepaid under current conditions to the deployment and use of PLC technology (Fig 510)
Smart grid development trends The European Commission by way of their DirectorateGeneral for Communications Networks Content and Technology in Brussels also believethat smart grids will play an important role in increasing the importance of renewable andalternative energy sources for low-energy consumption delivery savings and decreasingthe CO2 emission Without integration between telecommunication and informationnetworks the established goals are unattainable Smart grid is therefore a significant partof the long-term research and technology development program called Horizon 2020 [6]
The German Association of Electrical and Electronics Engineers VDE (in GermanldquoTechnisch-wissenschaftlicher Verband der Elektrotechnik und Elektronikrdquo) insists onplanned efforts for transforming of the traditional electricity networks and the creation
51 Smart Grid as Integration Technology for the Networks of Energy 149
of intelligent nets In several European countries this approach has become a significantpart of the national energy policy In this case it is not about some individual decisionsfor ldquoseveral thousand kilometers of cable or 100 million eurosrdquo Instead integratedsolutions for the smart grid must be developed during a middle-term period The mainobjective is as follows re-construction flexibility of the entire system re-design withelements of the modernisation of infrastructure increasing of capacity and number ofpower plants [18]
Meanwhile the approaches in the development of smart grid systems in the worldeconomy are very individual Let us consider some of them in detail
1 Australia The orientation to the development of intelligent energy supplying networksand smart grid has been taken in 2009ndash2010 WiMAX networks play an important rolein the frame of smart grids as a transport for support of applications for sub-stationautomation hybrid electric vehicles (PEV) as well as for domestic smart meters socalled IHD (In-Home Devices) However the final implementation of smart grids inAustralia is constrained by the lack of appropriate multilateral obligations between theproviders The inter-operability between the stakeholders has to be developed aimedto maintenance of communication networks that are integrated into the smart gridThe other limiting factor is a relatively small number of charging stations for electricvehicles despite obvious increases
2 China In the frame of the ldquocurrent five-year planrdquo for the Peoplersquos Republic of Chinaa construction of a national-wide monitoring system for national energy networkshas been started titled WAMS (Wide Area Monitoring System) The WAMS usesthe offered devices called PMU (Phasor Measurement Units) from selected Chinesemanufacturers to improve the reliability and security of the national smart gridsolutions Electrical energy production and distribution as well as broadband datachannels are tightly and restrictively controlled by the state Therefore complianceand conformity with existed standards and processes on the way of transition to anational smart grid is practically guaranteed There are already more than 60 millionsmart meters installed in China [44] although studies about the operations experienceare rare
3 South Korea The state plans until 2030 to reduce the overall consumption of conven-tional energy sources by 3 and electricity by 10 despite rising industrial demandsdue to the implementation of a nationwide smart grid The start has been taken in 2009the planned amount of investments for the system development for the next 20 yearsis about 24 1015 USD in equivalent to the national currency in South-Korean Won(KRW)
4 European Union The development of intelligent networks towards smart grid isa part of the European Technology Platform for the period up to 2020 devel-oped by CENELEC (in French ldquoComiteacute Europeacuteen de Normalisation Eacutelectrotech-niquerdquoEuropean Committee for Electrotechnical Standardisation) [4] The committeeCENELEC is occupied in charge of European standards in the field of electrical
150 5 Smart Grid Internet of Things and Fog Computing
engineering Together with ETSI (Telecommunications Standards Institute in the EU)the committee works on a European system of technical regulation and standardisationincluding the mentioned smart grid techniques models and tools
5 USA The support for smart grids became a part of the US federal policy towardlegislatively approved energy independence and security of one of the strongesteconomies in the world The amount of investment towards the middle-term develop-ment of this new technology will reach up to 11 trillion dollars ie 11 1012 USDaccording to plans from 2009 The short-term budget is however about 45 billiondollars according to the Recovery and Reinvestment act [17] Private microgrids arepart of the overall plan to turn the energy network into a bi-directional one similar tocommunication networks until the year 2030 On a global scale about 4000 megawattsare currently contributed by microgrids [36]
Example 55 An example of a connected smart grid and cloud computing implementationis given below Due to use of todayrsquos powerful high-end servers within the contemporarydata centres with the installed broadband optical links (so-called Fibre Channel) asignificant amount of heat stands out as a harmful by-product Some companies occupythemselves already with the mentioned problem and are developing their own solutionsfor the disposal of heat excesses for eg domestic heating and air-conditioning facilitiesor HVAC (Heating Ventilating and Air Conditioning) An imaginary joint-stock companyECO-Cloud is situated in a city of about 500000 to 1000000 habitants in Western Europeand acts as a data centre and cloud provider Several corresponding products and solutionsare offered cloud products (own virtualised data centre) and heat products (own smartgrid)
The temperatures of the servers can reach up to 55 degrees with water heat canalisationand dissipation The system of the waste heat recycling delivers a PUE of approximately106ndash115 Multiple clients use HVAC facilities in the city of the ECO-Cloud offices aswell as in other remote sites They could obtain up to 30 of cheap heat and warm waterfrom the mentioned clouds immediately The facility grids companies act as partners forthe ECO-Cloud with further 70 of the clients (users of the Internet standardised dataand cloud services) The waste heat distribution principle (based on [39]) is presentedin Fig 511 The company ECO-Cloud uses virtualisation technologies to create thecomputing storage and networking infrastructure The solutions are based on integratedcloud stacks as technology set
The clients use the in-door located services of virtual computing centres Hybridclouds with standard services spanning across company-internal and ECO-Cloud-hostedmachines are offered via ECO-Cloud too The IT resources such as operating sys-tem applications run-time platforms test and development environments as well aspurely processing power memory or network capacities and much more can be madeavailable to the users if necessary The computing centres encompass standardisedcloud services like Infrastructure-as-a-Service (IaaS) Software-as-a-Service (SaaS) and
51 Smart Grid as Integration Technology for the Networks of Energy 151
Fig 511 Waste heat distribution principle
Platform-as-a-Service (PaaS) as well as specific compute applications (compute serviceRAIDRAIC SAN NAS cloud stacks web hosting virtual operating systems file storageand sharing) [38] Redundant heat as a ldquoby-product of processingrdquo is withdrawn via serversin 19-racks into the energy storage which provides circulation of hot water in the pipeswithin a building and heating of potable water The central system for HVAC facilities issupported via use of PoE (Power over Ethernet) as well as wired and wireless automationLANs
While ECO-Cloud is an imaginary company nowadays multiple companies havespecialised on such business models An example is the former Helsinki electricity stationwhich still contributes to municipal heating due to diverting excess heat from the serversand racks installed in it nowadays The next two detailed examples will highlight additionalconcrete cases
Example 56 Similar principles are used by the high-tech company CloudampHeat Tech-nologies [39] The analog to the mentioned technical solution provides a lower PUE valuedown to 106 by the middle Tcpu D 55 ıC compared with the conventional grids and cloudsolutions where it is necessary to remove the excess heat as by-product to install more air-conditioning devices and provide them with power supply The construction of the waste
152 5 Smart Grid Internet of Things and Fog Computing
Fig 512 Redundant heat and energy recycling in the systems of smart gridcloud computing onthe example of CloudampHeat (Based on source wwwcloudandheatcom)
heat distribution can be depicted as in Fig 512 With such a construction up to 30 ofheat and warm water supply can be retrieved from the on-site cloud facility
Example 57 Another concrete example is IBH an innovative and customer-focusedcompany IBH provides the following services
1 Internet services including Internet access channels like SFV DSL MPLS or Metro-Ethernet
2 Hosting services for servers and complex IT installations as well as Application ServiceProviding (ASP)
3 Cloud computing services4 Domain registration and management as well as security certificates5 High-reliable fault-redundant three-phase Uninterruptable Power Supply (UPS) up to
4400 kVA
51 Smart Grid as Integration Technology for the Networks of Energy 153
Thus the waste heat from the data centre can be used for the heating of the buildingThe lost unusable waste heat from the computing centre is ecologically cooled via thedeployment of so-called ldquoindirect free coolingrdquo which enables an extraordinary highenergy efficiency for the computing centre ie a value of PUE lt 12 is attempted [8]
Example 58 Surely ldquogreenrdquo means a significant PUE improvement The ldquostate-of-the-artrdquo in a data centre today is to increase the temperatures in the server room gradually Thegeneric empirical ldquoformulardquo is in force see Eq 52
T D 1ıC H) PUE D 2ndash4 (52)
The formula expresses that an extra degree of heat gives 2ndash4 of energy efficiencyimprovement Energy efficiency improvement considers therefore its minimisation to thevalue PUE of about 10 with the same further decimal positions after the comma and thefirst zero
From formerly freezing air temperatures of T = 1112 degrees up to above 1617 degreesas the longtime standard for data centres the servers are nowadays being cooled down tothe rather warm level of 20ndash22 degrees without problems [31] With innovative solutionsindoor air temperatures are even increased up to 2324 degrees Very brave installationsare set up to go with the supply of air temperatures even higher than that The realitylies behind the technical possibility which means still far behind only 20 to 30 ofdata centre operators are already pursuing concepts and solutions for ldquohigh-pushing thetemperaturesrdquo [32]
Optimisation of cloud services for smart grids Google achieves a PUE of 112 due tofurther optimisation of hardware waste heat recycling systems and building constructionfeatures like improved air circulation reuse of waste heat and further techniques Thismeans that only 12 of energy required for computing is used not by servers as com-puting entities but by other services like air conditioning energy distribution lightingsurveillance systems and diverse building automation systems
Due to the ratio which is equal with and without consideration of time the PUE isdetermined as follows
PUE Dtotaldatacentreenergy
ITequipmentenergyD
totalpower
ITpower(53)
According to the Uptime Institutersquos Data Centre Surveys which track the average PUEin data centres by collecting survey responses there is a clear trend of reduction whichin the year 2007 was reported to be around 25 [19 40] The first survey in 2011 reportedan average PUE in the domain of about 189 As the fifth survey published in 2015 tellsthe PUE was reduced to 17 This means a significant improvement on the side of Googleeven though more than half of the data centre operators plan for a medium-term PUE of15 or less
154 5 Smart Grid Internet of Things and Fog Computing
Fig 513 Optimisation of cloud services for smart grids parallel computing and big data
The PUE thus becomes an attractive optimisation goal for service providers It affectsthe operational expenses whereas other optimisation targets focus more on capitalexpenses for the procurement of goods including the average server refresh rate whichcan be increased with high-quality hardware and good maintenance and repair servicesagain involving operational expenses Equation 54 formalises the operational goal aroundthe PUE optimisation
MaxPUE^
QoS Constraints_
Cost Constraints (54)
Where Costsmax QoSmin are the cost and quality of service constraints ie maximumPUE by strictly given QoS and cost constraints
In the third phase where we are now (maximum PUE by strictly given QoS and costconstraints) the following options of further improving the energy efficiency are attractiveand will most likely be used for contemporary data processing services (Fig 513)
1 Simultaneous operating of as few units as possible thanks to service and resourcevirtualisation increased resource sharing and load balancing
2 Better load utilisation of operating units eg by dynamic operation of serversdistribution of virtual machines and scheduling
51 Smart Grid as Integration Technology for the Networks of Energy 155
3 Using of more energy-efficient units (measured in Watt per GHz) to need less energyfor cooling
4 Optimised selection of location eg in cold regions close to rivers free cooling5 Reuse of waste heat eg for building heating or warming of potable water6 Use of a mix of local or regional energy producers to reduce transmission losses This
requires a smart energy grid and brokering ie a marketplace application in the cloudto work on a larger scale
Waste heat models To optimise the PUE it is essential to understand how to modelwaste heat and in particular the transport of waste heat The direction of transport isfrom the non-optimal computing equipment in particular CPUs acting as excess heatproducers to water or air as excess heat consumer media To understand the physicalbackground knowledge from the field of thermodynamics kinetics and green computingneeds to be combined Through more precise and fitting models the utility of smart gridswhich combine power systems and computing systems will be increased
The model will be derived from a state-of-the-art data centre perspective 9-inch-racksaccording to the norms EIA 310-DIEC 60297 are widely be used for the data centre andcluster construction The slots for such racks are called units or height units in jargonsimply 1 U One rack unit counts 175 inches (4445 mm) of height The following set ofunit dimensions H W D is wide-spread (Eq 55)
H D 17500 D 4445 mm D 1 U
W D 1900 D 482 6 mm
D D 600 800 900 mm
(55)
The 1900 rack containing the units has the following fixed dimensions The width W is19 inches (4826 mm) and gave the name to this standard The depth is derived directlyfrom the unitrsquos D The height H is determined by the industry standard for a rack cabinetwhich is 42 U and hence 4445 mm 42 D 18669 mm D 187 m These dimensions aretaken as input to a simplified Boltzmann waste heat transport model Excess heat recyclingand transport can be formulated and solved for the following constructions are given inFig 514 The shown principle of the removal and recycling of the energy can be usedfor the additional HVAC capacities within the civic administrative as well as industrybuildings
In the general case the Boltzmann model is linked to the Boltzmann ThermodynamicEquation (BTE) which for the heat balance can be given as specified in Eq 56
Pa D cmmmdTs
dtC PtI Pt D
SCTs Tw
RTI RT D
lmT
mT(56)
156 5 Smart Grid Internet of Things and Fog Computing
Fig 514 The waste heat recycling and transport principle (a) rack with units (b) unit with wasteheat removal device
In this equation Pa is the power absorbed by the system and Pt is the useless (excesswaste) power expended to the thermal conductivity Ts is the temperature of surface andTw the temperature of cooling liquid or cooling gas for example water cm is the thermalcapacity of heated materials and mm their corresponding mass RT expresses the thermalresistance of heated materials which depends on their temperature Finally m and lm referto the thermal conductivity and the thickneck of material respectively
Taken into account that for the stationary regime of heat exchanging the quotient of dTs
and dt becomes 0 the equation system can be rewritten as follows (Eq 57)
Pa D Pt DSCTs TwmT
lmT (57)
Based on the equation to build the waste heat model one now onsiders the complexthermodynamic problem of the cooling processors units as a task of simulation of a regularthermodynamic system In this system the sources of heat are named S Their squaresurface is similar and equal to a b and the distance between cooling units is named lAmong the length of cooling units a tube T with cooling liquid or gas is mounted Theheat is transferred along the tube with the velocity vc The corresponding model of thecooling system is plotted in Fig 515
51 Smart Grid as Integration Technology for the Networks of Energy 157
Fig 515 Generalised structure of cooling process for two processors unit (1) cooled downprocessor unit with the dimensions a b and the square Sc D ab (2) tube T with the coolingliquid or gas
The accuracy of the estimations for the temperature of crystal surface Ts temperatureof cooled liquid or gas Tw and of the power given by Eqs (56) (57) is not very high dueto multiple thermodynamic processes which act during the interaction between heated andcooled materials surfaces Those processes are not taken into account for the simplifiedexplanation In general the accuracy of such calculations is not greater than 30 Inany case these calculations for solving BTE can give the necessary recommendationsto engineers for elaboration and using of cooling systems For example the suitablerecommendations for the design of the cathode cooling systems for glow dischargeelectron guns were formulated and described as theoretic techniques in papers firstTherefore a similar approach for thermodynamic models computing is possible too Theuse of massive computing power for instance HPC allows for obtaining a higher accuracyin solving thermodynamic equations with finite elements
The presented BTE model can be decomposed into three subordinate models Themodel BTE1 is aimed at waste heat removal based on the cooling liquid or gas within thetube T in the area of a processor unit S The heat removal is carried out via a compoundadapter The second model BTE2 is dedicated to cooling down of the cooling liquidafter its heating in the tube T in the area between the units The length of this area is lcorrespondingly to Fig 515 Model BTE3 is the combined model of the models BTE1and BTE2 for a rack with N units
When solving the equations associated with the models BTE1 through BTE3 thefollowing observations can be drawn The PUE resulting from BTE1 and BTE2 is close to12 for both and about 13 for BTE3 These results match the state-of-the-art PUE factorsin data centres with standard cooling More details formulas and theoretic considerationscan be found in a relevant publication [34]
The estimations for temperature of CPU crystals (within the units) for temperatureof cooling liquid (waste heat removal) and the PUE evaluations based on the mentionedmodels have been considered in [25 33 43] The dependencies of PUE are given in
158 5 Smart Grid Internet of Things and Fog Computing
2
28
26
24
22
2
18
16
14
12
110 15 20 25 30 35 40 45 50
P = 500 WP = 400 WP = 300 WP = 200 WP = 100 W P = 500 W
P = 400 WP = 300 WP = 200 WP = 100 W
P = 500 WP = 400 WP = 300 WP = 200 WP = 100 W
PUE1 rel unit
PUEΣ rel unit
PUE2 rel unit
15
110 15 20 25 30 35 40
3
28
26
24
22
2
18
16
14
1210 15 20 25 30 35 40 45 50
45 50litermin
vel litermin
vel
litermin
vel
a b
c
Fig 516 Modelled PUE dependencies on the dissipated power of the processor units and givenvelocity of water flux based on the model BTE1 (a) BTE2 (b) BTE3 (c N=10) Modeled PUEvalues for ldquogreenrdquo data centres and clouds are about 106 under use of the efficient cooling processgood agreed to the ldquobest practicesrdquo
Fig 516andashc correspondingly These depictions describe the obtained PUE values for theabove-mentioned models BTE1ndash3 and are completely in line with the ldquobest practicesrdquodiscussed in the examples V4ndashV7 in this chapter as well as in the literature
Note Additional material on waste heat modelling and recycling is available ascomplementary digital-only material from the publisherrsquos website
512 Smart Grid Enabling Network Technologies
Enabling networking and communication technologies for smart grids offer wirelessconnectivity between devices Six such technologies are of particular interest
1 PLC outdoor as well as indoor as homeplug2 Bluetooth v42 WPAN
51 Smart Grid as Integration Technology for the Networks of Energy 159
3 ZigBeeEnOcean sensor piconets4 6LoWPAN as fog computing predecessor5 WiMAX networks specific to some regions with sufficient coverage6 Partially LTE5G which are discussed in other chapters as alterantive to WiMAX
Let us discuss some of them to understand their characteristics better
Powerline PLC PLC networks are oriented to use electrical supply networks (grids)for data and voice transfer This is an important enabling technology for IoT and smartgrids The network transmits data or voice by superposition of an analog signal over thealternating electric current (AC 5060 Hz) PLC in the WAN area offer a kind of DSLconnection via a power cable between providers and users
bull 1536 subcarriers with 84 best frequencies in the range 2 34 MHzbull Data rate per station of about 15ndash205 Mbitsbull Variants of WAN PLC are BPL and NPL NPL (Narrowband over Power Lines) with
data rate of 15 Mbits and BPL (Broadband over Power Lines) with data rate of205 MBits
PLC in the LAN area are more suited to applications within buildings PowerLAN presentshousehold electrical lines with a voltage of 230 V and a frequency of 5060 Hz foradditional data transmission
For such installations Orthogonal Frequency-Division Multiplexing (OFDM) isdeployed for converting digital signals into analog signals similar to XDSL or WLANMost PowerLAN standards work in the high frequency band F D 2 68 MHz so thatthey do not interfere with the electric current frequency and with the aim of achievinghigh data rates Power supply networks as low voltage networks are usually three-phasesystems In the private sector the HomePlug standard thus achieves gross transfer rates ofup to 14 Mbits (regular HomePlug) 85 MBits (HomePlug Turbo) 200 Mbits (HomePlugAV) and even 500 Mbits (IEEE 1901) The standards HomePlug AV (200 Mbits) andIEEE 1901 (500 Mbits) are fully compatible with each other The maximum range ofHomePlug adapters is however limited to 300 m under ideal conditions and much lesswhen obstacles are in the way
The PLC usage main problems and violations are
bull line lengthbull interferencebull interoperabilitybull price
Longer lines mean the occurrence of attenuation effects which limit the transmission powerand hence reduce the receiverrsquos ability to process the signals effectively leading to a
160 5 Smart Grid Internet of Things and Fog Computing
Fig 517 WiMAX flexible architecture
reduced data rate The interference comes from the workload and household machinesThe interoperability with Wi-Fi is not guaranteed as producers are unable to agree on acommon standard Finally such systems are still subject to a relatively high price
WiMAX networks The architecture components for WiMAX networks are depicted inFig 517 Among the WiMAX components are
bull SSMS Subscriber StationMobile Stationbull ASN Access Service Networkbull BS Base Station a part of ASNbull ASN-GW ASN Gateway a part of ASNbull CSN Connectivity Service Networkbull HA Home Agent a part of CSNbull NAP Network Access Providerbull NSP Network Service Providerbull ASP Access Service Provider (IP)
The most important interfaces are R1 R2 R3 R4 R5 (refer Fig 517) The use of WiMAXis regional specific It is frequently used in South Korea South Africa (named iBurst)
51 Smart Grid as Integration Technology for the Networks of Energy 161
and the Slovak Republic as well as in urban areas in other countries An example isHeidelberg in Germany But commonly the WiMAX networks found however a relativelysmall acceptance compared with LTE In fact many former deployments have been shutdown already for instance by Sprint in the USA Still about one billion people can becovered
The maximum distance for signal transmission is about 3ndash10 km
Sensor pico nets As opposed to the previously discussed network types which emphasisequality of service and cost requirements wireless sensor (pico) networks (WS(P)N)additionally put emphasis on various aspects of energy efficiency A WSNrsquos energyefficiency is a significant prerequisite for its lifetime low maintenance cost and highreliability First a short overview about WSN systems will be given Then the mostimportant compromises or trade-offs between the diverse factors will be discussedespecially those which influence energy efficiency and service quality on any networklayer
WSNs have already become a mature technology and play an increasingly importantrole for industrial production intelligent houses automated buildings and observationin the free space in agriculture and forestry ecology and ship transport This list ofapplications of WSNs is however far from being complete Advanced WSNs replace incombination with WLAN and WiMAX networks conventional communication systems formulti-function network services and automation systems
A general sensor network consists of a number of distributed and independent sensornodes (SN) with radio modules These are capable of capturing technical or environmentalparameters There are many different sensor types and technologies of which two shallbe considered (Table 52) Common to all these technologies is the issue of energy-efficient operation of the resulting sensor networks Energy-efficient sensor nodes arecharacterised by durability interoperability and assurance of quality of service levels(QoS) within constructed WSNs Furthermore they are highly reliable and contain cost-efficient customisation mechanisms
Table 52 Characteristics ofwidely-used WSN systems
Property EnOcean ZigBee 802154
Frequency MHz 868 2400
MAC layer Beacon Beacon CSMA
Topology Starmesh Starmesh
Data rate KBits 125 250
Number of nodes 232 D ca 4 milliards 216 D 65536
Security ndash AES
Energy consumption Very small Small
Collision probability Very small Small
Energy harvesting Yes No
Range m 30ndash300 10ndash75
162 5 Smart Grid Internet of Things and Fog Computing
Fig 518 Structure of a WSN
The usual frequency bands F for WSN are F D 315 916 MHz (Mica2 Mica2Dot)and F D 24 GHz (ZigBee IEEE 802154 Imote) The usual transmission ranges ofsensor nodes can be from 30 up to 150 m The energy consumption is about 1000 mW forsending and receiving data 100 mW in idle mode and 005 W in sleep mode The averagetransmission power is PTx D 4 10 dBm To guarantee the requirements concerningenergy efficiency and real-time behaviour only short data packets (telegrams TL
100 bytes) with relatively small overhead are being used The state transition of a sensornode (SN) requires energy and slows down the network overall
The approach of energy harvesting allows for the extraction of energy from theenvironment and thus for a reduction of battery power consumption (Fig 518) Theexclusive energy supply of sensor nodes with energy harvesting is however not possibledue to the lack of steadiness in the used energy sources Therefore the nodes have to beplaced with care Furthermore an optimisation of routes to the gateway (GSNGW) isrecommended
The software used on the nodes (operating system applications libraries middleware)has to be very compact The executed tasks and the data to be processed often have to
51 Smart Grid as Integration Technology for the Networks of Energy 163
be scheduled preliminary and grouped with telegram aggregation For the minimisationof the energy consumption of the communication (SN ndash SN and SN ndash GW) and forincreasing the performance of the gateway concepts such as caching threading andredundancyreplication are to be considered The task processing in the applications isevent-based [45] As operating system for the sensor nodes Tiny OS is often used It hassmall requirements on memory and processing power
Design of energy-efficient wireless sensor networks Requirements and methodsImportant properties of energy-efficient WSNs are
bull Efficient batteries with long lifetime in the sensor nodes possibly combined withenergy harvesting
bull Energy managementbull Efficient protocols in the layers 2 and 3 with reduced traffic and low overheadbull Efficient operating systems and applicationsbull Optimised topology including hierarchies and clusteringbull Redundant planning and functionality reservesbull Combined approaches in a cross-layer design
Multi-layered design Nowadays the design of WSNs is supported with a variety ofenergy management methods and planning tools The cross-layer approach combinesexisting models methods and tools within one integrated framework and offers significantadvantages due to the holistic appreciation of values between requirements of energyefficiency and service level The methods for designing energy-efficient WSNs can beclassified in a layered architecture as follows
bull Hardware focusing on the physical (PHY) layerbull Focusing on the MAC layerbull Focusing on the topologybull Focusing on routingbull Focusing on applications
An attempt for a corresponding classification of methods usable for the design of energy-efficient WSNs is shown in Fig 519
Efficient energy management for WSNs primarily means that the overall powerconsumption of a WSN must be reduced by optimising the consumption of its sensor nodesexpressed in Wbit or Wevent Such an optimisation leads to an extension of parameterswhich indicate the lifetime (time-to-live TTL) expressed in 1000 h or 100 d The followingparameters are common T1 ndash time until the failure of the first sensor node T2 ndash time untilwhich 50 of all nodes fail T3 ndash time at which the network splits into multiple partitionsor ldquoislandsrdquo T4 ndash time until the surface coverage of the network is reduced The TTLparameters are explained in Eq 58
164 5 Smart Grid Internet of Things and Fog Computing
Fig 519 Classification of design methods for energy-efficient WSNs
The cross-layer construction of WSNs needs to consider the mutual influence of theconflicting requirements energy efficiency and service level Appropriate compromisesneed to be found
bull Hardwarendash Higher transmission frequency more data per TDMA slot as well as more compact
components but more complex modulation techniques and higher energy consump-tion requirements
ndash Lower transmission power less energy consumption upon transmission but lowersignal-noise ratio (SNR) and lower data throughput
ndash Lower current of the components (cf Fig 520) lower energy consumption of theCPU but correspondingly lower CPU speed
ndash Higher battery capacity longer lifetime but larger physical dimensions This isalso true for energy harvesting approaches which require sufficiently strong energysources and batteries in order to adjust the non-continuous energy supply
bull MAC layerndash Longer sensor duty cycles in communication protocols (eg synchronous
on-demand TDMA or Advanced Asynchronous CSMACA with RTSCTS orRendezvous) improved degree of utilisation but also higher latencies
51 Smart Grid as Integration Technology for the Networks of Energy 165
Fig 520 (a) Dynamic voltage scaling (b) Capacity of batteries and energy harvesting devicesApproaches to optimise the energy consumption
bull Topologyndash Cluster of nodes following a unified scheduling scheme with lower duty cycle lower
power consumption in sensor nodes through shorter distances but higher latenciesthrough overhead and higher energy consumption at the cluster head
ndash Dense WSN with redundant nodes higher availability and reliability but alsoincreased traffic and therefore more collisions of data telegrams as well as morefrequent timeouts
bull Routingndash Highly developed routing algorithms (eg geographic routing) increase the reliabil-
ity of the message transfer but cause higher routing complexity and therefore morelaborious routing adaptations in cases of topology changes
bull SoftwareApplicationsndash Compact operating system and further software components due to limited CPU
speed and RAM capacity better resource utilisation but lower precision throughdata aggregation as well as a necessity for special algorithms for distributedstatistical pre-processing of large volumes of data
166 5 Smart Grid Internet of Things and Fog Computing
These compromises (trade-offs) need to be accounted for in the design phase to achievethe goal of durable WSNs with high QoS high reliability and interoperability betweenthe nodes The stored energy density can vary between 10 and 10000 Wcm3 Thedetermination of TTL parameters can be performed by considering the following factors
minTTL D ˛ıq
ıxPTx F d DR SNR TL OH (58)
Hereby q refers to the battery charge [mAh] F and PTx to transmission frequency andpower d to the average distance between nodes (hop distance) DR to the data rate TL tothe average size of a data telegram and OH to the overhead in each data telegram is acentralised Gaussian random value whereas ˛ is a logarithmic decrement value
Topology optimisation The most important decision when designing topologies of aWSN is the choice between single-hop and multi-hop routing methods
The following aspects are to be considered Who communicates with whom (starcluster or mesh) incomplete knowledge about the topology only information about thelocal environment is known frequent topology changes on-offboarding mobility aspectsrouting algorithms and of course the energy efficiency of the resulting solution
The degree of freedom for the decision can be described as a triangle ldquotopologyndash routing ndash energy radiationrdquo which is displayed in Fig 521 The power radiation ismodelled as follows
PRx D KF˛d K D PRxdref (59)
Whereas PRx refers to the receiver field force F to the sender frequency d to thedistance and PRxdref dref to the measurable reference receiver power and distanceK ˛ are model constants from the free space damping model
Clustering in WSN When nodes of a WSN are distributed in fixed installations thehardware will degrade over time After some years some of the nodes may fail or thebattery capacity may be depleted In such cases it is important to consider the correctplacement of the nodes to avoid missing hops for the transmission or even partitionednetworks in which between any two nodes one from either partition any communication ispossible Failures and ldquodesertificationrdquo effects are depicted in Fig 522 Optimal clusteringand a certain amount of transmission link redundancy is therefore required
LEACH description Low-Energy Adaptive Clustering Hierarchy (LEACH) is analgorithm which clusters nodes so that the communication between any two nodes orbetween any node and a base station is routed through cluster heads The nodes that werealready cluster heads (CHs) cannot play role of CHs for next 1
p rounds where p is thedesired percentage of cluster heads in the network Furthermore each node possessessome probability Z lt Tn to become the cluster head in a new round At the end of
51 Smart Grid as Integration Technology for the Networks of Energy 167
Fig 521 Topology ndash routing ndash energy radiation Energy efficiency via topology and routing
Fig 522 Failures and ldquodesertificationrdquo effects [37]
168 5 Smart Grid Internet of Things and Fog Computing
the round each of the nodes which have not become head calls the next CH and becomesa only cluster member (Join Cluster) Then each of the CHs have to establish a plan (clusterschedule) for each node This enables a successful data transfer for its own cluster
bull Spatially distributed applications with data aggregationbull Cluster Heads (CH) are defined locally and randomisedbull They have to be periodically replacedbull Energy efficiency
Figure 523 shows the LEACH algorithm in an example to increase the lifetime of piconets
Sensor piconets ZigBee and EnOcean Both technologies ZigBee and EnOcean areenablers for smart grids and important for IoT and fog computing Their characteristicswere already identified earlier (remember Table 52) They found their usage for intelligenthome process control robotics automotive and aviation The components are sensors vsactuators (servomotors pumps heating controls) The typical bottlenecks are batteriesaccumulators privacy and anonymity Due to a limited amount of energy they havetherefore less reliability and more expensive maintenance This is the reason why energyharvesting ie use of ambient energy (solar EM smog noise ) is a very importantoption
bull solar radiation consider during the installationbull reduction of maintenance costsbull cheaper materialsbuilding materials
WPAN ZigBee The name ldquoZigBeerdquo derives from the zig-zag dance of the bees by foodsearching ndash in analogy to the traffic in a meshed network ZigBee is designed as a WPANeffectively a low-data rate PAN and uses IEEE 802154 specifications for the PHY andMAC layer as shown in Fig 524
A short history of ZigBee systems
bull 1998 ndash ZigBee launched by Philipsbull 2001 ndash IEEE 802154 based ZigBee Group foundedbull 2002 ndash ZigBee Alliance founded (Philips Mitsubishi and 230 other companies)bull 2005 ndash first ZigBee products appaeared on the marketbull 2007 ndash current standards ZigBee 2007 release
ZigBee products fully conform to the requirements of low-rate wireless PANs with thefollowing features
bull low data ratebull long battery life
51 Smart Grid as Integration Technology for the Networks of Energy 169
Y Y
X X
Surviving nodes []
Life durability [days]
14000
100
50
700 1050350
Direct connected Static clustersLEACH
a
b
Fig 523 (a) Different LEACH cluster heads in the neigbour rounds the round 1 and round 2 (b)Clustering by LEACH better surviving LEACH approach to clusters head asssingment long lifepiconets [37]
bull secure networking with AES encryption and WPA2 authentication
There are three roles for ZigBee devices
bull ZigBee End Device (ZED)bull ZigBee Router (ZR)bull ZigBee Coordinator (ZC)
170 5 Smart Grid Internet of Things and Fog Computing
Fig 524 ZigBee layer model(Own representation)
The ZigBee End Device (ZED) is a simple device such as a light control It implementsonly part of the ZigBee protocols and is therefore also called RFD (Reduced FunctionDevice) One is logged on to a router of their own choice then they form a star topologywith it The ZigBee Router (ZR) refers to FFD devices which can act as routers Onecan log on to an existing router by forming a tree or mesh topology Finally the ZigBeeCoordinator (ZC) is a special router within a PAN It takes the role of coordinator Thus itcontrols the basic parameters of the PAN and manages the network The general topologyof ZigBee systems is shown in Fig 525
ZigBee systems operate in the ISM band with a frequency of F D 24 GHz and datarates of DR D 025 MBit=s for a range of 10ndash75 m In the MAC layer either CSMACAis implemented or so-called Beacon signals are sent similar to how a lighthouse worksThe Beacon signals are sent by a cooperating station after longer communication idlenessintervals All network participants within the proximity will become ready to receive for acertain amount of time Collisions become unlikely with this technique
ZigBee offers compatibility to alternative solutions on the layers 1 and 2
bull USA and China ndash 902915 MHz 40 kBitsbull Japan ndash 928 MHzbull Other Asian countries ndash 315 MHzbull Europe ndash 868 MHz 20 kBits
51 Smart Grid as Integration Technology for the Networks of Energy 171
Fig 525 (a) Star vs P2P (b) Multi-hops ZigBee topologies
However more possible interferences with existing WLAN networks need to beconsidered The most important applications of ZigBee products are
bull Structural Health Monitoringbull Facility Managementbull Smart Metering usw
The next rival is EnOcean
WPAN EnOcean The company EnOcean located in Oberhaching near Munich belongsto Siemens EnOcean a system of wireless sensors with power self-supply by energyharvesting is broadly used in the area of building automation They are similar althoughalso distinguishable from ZigBee systems as shown in Fig 526
172 5 Smart Grid Internet of Things and Fog Computing
Fig 526 Sensor piconets ZigBee and EnOcean in comparison CO2 reduction from Airbus planeswith sensors and home automation with thousands of sensors in the Torre Espacio in Madrid a56-floors building (Sources airbuscom eswikipediaorg)
EnOcean offers a high energy efficiency by combining the transformation of locallyavailable environmental energy with dynamic voltage scaling and very short duty cyclesEnOcean systems have been practically known since the year 2001 In 2008 the EnOceanAlliance has emerged from several well-known companies from multiple countries (DEFR EU USA) among them Siemens and Osram In 2015 EnOcean focuses on buildingautomation with several products switches sensors receivers and controllers gatewaysmanagement systems and accessories Furthermore there is a joint development withZigBee 30 for energy harvesting
EnOcean products work over distances from 10 to 300 m For the design of EnOceansystems an optimised cross-layer approach is followed (Fig 527) The MAC layer isbased on beaconing The associated collision probability is however relatively small Tominimise its effects pseudo-random short telegrams with a message length of 14 bytesare submitted three times The systems use the frequency band of F D 868 MHz andoffer low data rates with DR D 125 kBit=s However EnOcean structures are robust andenergy-conserving
There may be interferences to the following radio networks
bull GSM DECT ndash rare occasionsbull ZigBee 802154 ndash needs to be accounted for
51 Smart Grid as Integration Technology for the Networks of Energy 173
The use of EnOcean products happens through more than 50 system integrators whodevelop and produce products for the building automation (light shadows heating climateand air conditioning) industry automation and the automotive sector These systems aretypically more economical than their rivals and are broadly supported on the market forinstance in Germany France and other EU countries One disadvantage of the technologyin comparison with other WSNs is a lack of integrated security mechanisms
EnOcean is a good example for the compromises needed for the design of WSNs Thefollowing design criteria have been set to adapt to the low energy supply generated byenergy harvesting
bull Single hop to the cluster head flooding between cluster heads data processing incluster heads
bull MAC layer no collision detection but beaconing uni-directional communicationbetween sensors and cluster heads
bull Limited energy supply short telegrams (1 ms) and duty cycle (01ndash1 )
The EnOcean layer model is depicted in Fig 527 The main distinguishing features ofthese piconets in general are
bull low data ratebull long battery lifebull secure networking
Fig 527 EnOcean layermodel
174 5 Smart Grid Internet of Things and Fog Computing
They are analogous to ZigBee features but implement energy harvesting as uniquestrength ie incorporate the use of ambient energy and primarily solar (also EM smognoise )
Typically tasks of designing efficient and high-quality WSN deployments are
bull Energy-efficient protocolsbull Cross-layered optimisationbull Trade-offs between layers are to consider
The following layers are of interest
bull Hardware or PHY basedbull MAC basedbull Topology basedbull Routing basedbull Application and data basedbull Cross-Layered (combined approach)
Example 59 Think of a ldquotoy smart gridrdquo An example of a model environment for smartgrid (Smart Grid Simulator) [13 37] is presented in Fig 528 The modeling environmentconsists of a miniature city (eg is based on the famous German model railway toyldquoModelleisenbahnrdquo) The structures of the model cities are the buildings H1 H2 H3H4 a plant and a McDonalds restaurant all of which are placed on a portable board or atable
The emulation of ldquocustomersrdquo and ldquosuppliersrdquo of electricity is based on microproces-sors or single-board microcomputers AVR Raspberry Pi and Intel Edison are represen-tative products in this category The compact dimensions and low power consumption areamong the main priorities of on-board computers (see Tables 53 and 54)
Let us discuss the computing nodes based on Raspberry Pi [14] These computationalnodes are combined to a local area network (LAN) with low dimensions Each nodeRasPi1 RasPi2 RasPi3 operates one ldquobuildingrdquo and visualises on the display orLEDs LED1 LED2 LED3 the active ldquoconsumersrdquo and ldquosuppliersrdquo of electricitywithin the ldquobuildingsrdquo and in the system in general
With use of the ventilators and LED lamps the main ldquoweather conditionsrdquo like sunradiation and wind are emulated The modeling environment (so-called simulator) iscontrolled by the developed software scripts (running as WWW applications) and shouldmap to the changes of connections through reflection of the new ldquoconsumersrdquo andldquosuppliersrdquo of electricity as well as undertake the representation of some changes withinthe weather conditions Thus using the model environment within the artificial toy system
51 Smart Grid as Integration Technology for the Networks of Energy 175
Sensor
House 2
House 1
McDonalds
USB hub
Ethernet switch
House 3 Plant
D
D
D
D
D
D
RailwayStation
KB
KB
KB
KB
KB
LegendX Raspberry Pi
KB Keyboard
D Display
House 4
LED
Ventilator
KB
USBsupply
InterconnectionsGPIO
EthernetUSB
Wind
Weather
a
b
Fig 528 Example of a modeling environment for smart grid [13] (Photo nlwikipediaorgtopology inspiration rninftu-dresdende)
the real parameters and laquosmart gridraquo conditions can be modeled This includes the usageof intelligent network services electricity grids as well as the energy-efficient informationservices
176 5 Smart Grid Internet of Things and Fog Computing
Table 53 The distinguishingfeatures of on-board computers
Characteristics On-board computer
CPU type ARM Cortex Intel
GPU type Mali Intel PowerVR etc
RAM 05 up to 8 GByte
Price Approx 15 up to 100 $
Dimensions Max 2 5 cm
Power consumption 25ndash5 W
Table 54 Comparison of the chips and microcomputers AVR Raspberry Pi Intel Edison
Parameters AVR32 Raspberry Pi Intel Edison
Manufacturer Atmel CA 2006 CambridgeRaspberryPi Foundation UK2011
Intel CA 2014
Dimensions Middle Small like a bank plas-tic card
Tiny like an SD storagecard
Type RISC-CPU low power32 bit -controller
ARM on-board -computer
On-board -computer2-Core i-Quark 22 nm-transistor technology
Frequency 66ndash200 MHz 700 MHz 400 MHz
RAM Flash D 512 KByteRAM D 64 KByte
SD card instead ofHDD RAM 256MByte
ndash
Ports networkinterfaces
USB 20 serial USART 1x LAN Ethernet10100 RJ45 2x USB30 1x SD 1x HDMI1x ClincTRS adapter6x GPIO
Wi-Fi Bluetooth
Operating sys-tem
Linux Linux BSD UNIXRISC OS
Linux
Look
Board or pod ndash
approximateprice
20 19ndash30 ndash
51 Smart Grid as Integration Technology for the Networks of Energy 177
513 Case Study A CAD Toolset for the Design of Energy-EfficientCombined Networks
There are multiple tools which aid in the design of communication networks in particularsensor networks energy grids or combined smart grids In the following the tool CANDY(Computer-Aided Network Design Utility) will be introduced briefly Further literatureabout CANDY is available [27 29]
Basics on CANDY The energy-efficient combined networks in the context of smart gridscan be designed with use of CANDY Framework and Online Platform [27] We would likefurthermore to discuss important development trends for a CAD for combined networkplanning regarding to the tool integration and access The CANDY Framework and OnlinePlatform is examined as a reference system The CANDY system has been represented asan exhibit at CeBIT 2007 2008 2011 in Hannover Germany and has demonstrated itsusefulness for academic and industrial network planning challenges
A CAD toolset for combined office communication and building automation networks(sketched in Fig 529) is presented It especially focuses on the combination of wired
Fig 529 LON ndash Local Operating Network KNX ndash European Standardised Bus AutomationNetwork (EN 50090 ISOIEC 14543) PDA ndash Personal Digital Appliance ERP ndash EnterpriseResource Planning EDP ndash Electronic Data Processing A combined office communication andbuilding automation network
178 5 Smart Grid Internet of Things and Fog Computing
(IEEE 8023-LAN) and wireless (IEEE 80211-WLAN 80216-WiMAX) networks as wellas on wireless sensor networks using 802154EnOcean
The CANDY framework supports an integrated design methodology providing a com-plete design workflow The design requirements on these networks are often contradictiveand often have to consider diverse technical factors among them performance energy andcost efficiency for a network solution altogether
The system provides the following features
bull integrated workflow managementbull dedicated network description via NDMLbull structured cabling by EN 50173 supportbull front-end to CAD conformity (ifcXML) IP infrastructure analysisbull access services to a high-performance computer clusterbull as well as parallelised design routines realisation (OpenMP) [29]
Dedicated network language The framework uses the dedicated Network Design Mark-up Language (NDML) an XML-based notation to express modelled networks NDMLsupports a uniform way of representing all major active and passive network elements(including switches routers gateways patch fields cross panels base stations sensorsaccess points as well as automation nodes) their detailed technical properties as wellas their interconnections and related configuration issues In contrast to existing vendor-specific notations NDML is based on open standards and enables interoperability andportability of network design tools and projects
Tool integration concepts and access CANDY is an open framework with a large setof design tools and functionalities These include design editors consistency checkstransformation tools specific wireless network design tools and integration of existingsimulation environments NDML serves as common ldquogluerdquo for these tools Java tech-nologies facilitate the tool development including among others Application Server andMiddleware (Apache Tomcat with JSP Java Server Pages and EJB Enterprise JavaBeans) ERCP (Eclipse Rich Client Platform) as well as web services (Apache Axis 2)A flexible tool access is provided via available Java desktop applications and Androidapplications on mobile devices such as smartphones and tablets
Development history The CANDY tools have been developed along with emergingnetwork trends They went through the following development history
1 Conception and implementation of a prototype (CANDY Prototype)
bull Conception of NDML with prototype for network editorbull Prevalent implementation basis Java servlets Java applets EJB
51 Smart Grid as Integration Technology for the Networks of Energy 179
2 Realisation of dedicated planning tools (CANDY Framework) inter alia tools for
bull structured cabling system called CANDY Trace Routerbull optimised design of radio networks called CANDY Site Finderbull prevalent implementation basis Eclipse Rich Client Platformbull further development of NDML (XSD instead of DTD achievements in advancing
of viewpoints und language elements)bull realisation of an extensible framework (CANDY Framework) with mostly important
planning steps and front-ends to capsulated external tools
3 Further realisation of a universal design platform (CANDY Framework with CANDYOnline Platform)
bull workflow and documentation management (ldquoWF-centricrdquo)bull support of all design stepsbull loose embedding of capsulated external tools via web servicesbull prevalent implementation basis HTML5 AJAX web servicesbull creation of multiple agile mini-tools for combined network designbull multimodal access via mobile users with smart phones and tablets (cp Fig 530)
CANDY Frameworkand Online Platform
Modules1 ndash Project Manager
2 ndash Network Editor3 ndash Component Browser
4 ndash SCS Trace Router5 ndash Wireless Site Finder6 ndash Workload Analyser
7 ndash Bill ReporterFE ndash Front-end (XML)
T ndash Loose-coupled and 3rd party toolslike for instance NS 2
High-performancecomputing environment
DB ndash Component repository
Access viaCANDY Web
Services furtherinterfaces
FE
1
2
3
4
5
6
7
DB
Project Data
Component List Network List PerformanceReport
Cost Bill
NDML
Environment
T
T
ApplicationServer JREEclipse RCP
Fig 530 Design tool integration and access
180 5 Smart Grid Internet of Things and Fog Computing
After multiple iterations of development the system now possesses the followinghighlights which make it suitable for future networks and smart grids
1 Accurate planning is the precondition to decisive advantage under competition pressureIn view of networks complexity the task can be solved by use of efficient software toolslike CANDY Framework and Online Platform
2 Network engineers have to optimise large-scaled objectives within complex contextsCANDY represents an integrated design for 80238021180216802154 networksunder use of its own models as important integration component
3 The implemented CANDY Online Platform provides possibility to running of complexparallelised propagation algorithms for wireless networks as well as multi-variantTCPIP simulation processes in high-performance computing environment Thisdeployment mode was verified on MARS (ZIHTUD)
4 The realised framework and access services offer to the specialists and students a rarepossibility to start their ambitious CAD jobs obtain the results in few minutes supportreal measure data acquisition and their comparison with modelled results
Workflow-centric management A CANDY workflow for network design andldquoWF-centric managementrdquo are built under use of the following principles
1 A CANDY workflow is combined from a sequence of design steps2 Each step consists of one process (task) or multiple parallel processes3 Each process possesses a status eg (ready [yn] result [C=])4 Each process uses andor produces inputoutput documents5 A process is either an atomic process or a workflow by itself as shown in Fig 531
Simulation and validation The design results for WLAN IEEE 80211 are in general notsatisfyingly accurate Correspondingly a site survey functionality with design correctionis necessary for each installation (cp Fig 532a) An advanced method for the planning ofradio networks leans on the prognosis of the received power PRx and a comparison of mea-sure values aimed at their further optimisation The method is called ldquoMeasurement-basedPredictionrdquo (MbP methodology) The reference components of the MbP methodologyare shown in Fig 532 By deployment of the MbP methodology advanced measuredevices and hardware solutions can be used The databases contain all necessary referencevalues covering samples antenna coordinates and other metrics The used empirical radiopropagation model is valuated and via inset of the MbP methodology is adapted to the realreceived power PRx
An example of the practical use of mini-tools for the design of a wireless networkconstellation within the CANDY Online Platform is given in Fig 533
The discussed design steps within CANDY are furthermore presented in summary inFig 534andashi The design process starts with a topology editor (a) which outputs the basic
51 Smart Grid as Integration Technology for the Networks of Energy 181
Fig 531 WF-centric management
network elements and connections between them From the visual modelling a textualnetwork description (b) in NDML is then derived This description is then importedinto another modelling tool (c) and applied to a concrete deployment site for instancea building with an ifcXML description (d) The wireless and wired connections are then
182 5 Smart Grid Internet of Things and Fog Computing
Fig 532 Simulation and validation
Fig 533 Simulation via mini-tools within the CANDY Online Platform
51 Smart Grid as Integration Technology for the Networks of Energy 183
acce
ss to
a c
ompu
ter c
lust
er
star
ting
a re
mot
e jo
b
pick
up o
f the
resu
lts
from
com
putin
g cl
uste
r
a w
ired
part
SC
S
traci
ng fo
r Eth
erne
t LA
N
perfo
rman
ce s
imul
atio
na
NM
DL
repo
rt
a C
AD
-con
form
ifcX
ML
desc
riptio
n of
a b
uild
ing
a w
irele
ss p
art
envi
ronm
ent a
ttenu
atio
n
topo
logy
edi
tor
ifcX
ML
data
impo
rtnetw
ork
desc
riptio
n vi
a N
DM
L
a cf
gi
bd
eh
Fig
53
4D
esig
nro
utin
esm
odel
san
dto
ols
(a)
Topo
logy
edito
r(b
)N
etw
ork
desc
ript
ion
via
ND
ML
(c)
ifcX
ML
data
impo
rt(
d)A
CA
D-c
onfir
mif
cXM
Lde
scri
ptio
nof
abu
ildin
g(e
)A
wir
edpa
rt
SCS
trac
ing
for
Eth
erne
tL
AN
(f
)A
wir
eles
spa
rt
envi
ronm
ent
atte
nuat
ion
(g)
perf
orm
ance
sim
ulat
ion
aN
MD
Lre
port
(h)
acce
ssto
aco
mpu
ter
clus
ter
star
ting
are
mot
ejo
b(i
)pi
ckup
ofth
ere
sults
from
MA
RS
mul
ti-co
resu
perc
ompu
ter
184 5 Smart Grid Internet of Things and Fog Computing
tested and traced according to their specific characteristics (e f) Using the refined NDMLdescription a first performance forecast can be generated (g) Due to the complexityof modern installations with hundreds of network elements the simulation and othercalculations are best outsourced to a high-performance compute service as a remote job(h i)
In summary CANDY shows that not only the runtime and operational perspectivebut also the systematic and tool-supported planning beforehand is an important element inachieving high-quality network installations for basic connectivity cloud network servicessmart grids and connected things
52 From Internet of Services to Internet of Things Fog Computing
It may appear to the reader that we told everything about the properties of IoS and cloudcomputing today to the fullest satisfaction in the first chapters But there is another trendthat is impossible not to be highlighted within this book that claims to convey a broadscientific novelty Let us examine these new trends in network services collectively calledIoT as well as the ways of their realisation in the form of Fog Computing
The interconnection of scientific and technical ideas on Internet of Things Internet ofServices clouds and smart grids is shown in Fig 535 The mentioned technologies andtrends IoT IoT clouds and smart grids are developed in close cooperation and relatedto each other The presented organigram depicts additionally the approximate dates of
Internet ofServices (loS)
CloudComputing
2005-2010Amazone MS
Smart Grid
1999Auto-IDMIT
Kevin Ashton CiscoSAP Telefoacutenica
2011IEEE CENELEC Cisco
Deutsche Telekom Siemens
Internet ofThings (loT)
2004-2007WWW OASIS Google
Fig 535 Ideas development concerning IoT IoS clouds and smart grids
52 From Internet of Services to Internet of Things Fog Computing 185
the inset of the mentioned terms and categories as well appropriate system exmaples oftheir use with specifications on which organisations and companies are interested in thisdevelopment
Internet of Things The so-called IoT provides the radio-communication between multi-ple milliards of low-power devices within near distance up to global scale using protocolssuch as IPv6 The Internet of Services with its realisation in the form of clouds and withthe number of devices approaching N Š 109 nowadays will be shifted in the midterm toIoT The following distinguishing features are typical for this transformation
bull huge number of devices N gt 300 109 (probably after 2020)bull low power consumption and long-life nodesbull energy-efficent and secured communication radio-protocols interfaced to ldquonear fieldrdquo
and IPv6bull wide deployment within embedded systems and industry (cf Industry 40)bull penetration to each sphere of human activities and everyday life (Fig 536)
Highly concentrated deployments of connected things exist in South Korea Denmark andSwitzerland each having about 30 devices online per 100 persons according to OECD[12] This statistics apparently excludes interaction devices such as smartphones tabletsand notebooks which would significantly increase the statistics The IoT field overlapswith application areas for instance robotics smart cities transportation (through e-ticketsand on-board units in electronic toll areas) agriculture and environmental sensing
The origins of IoT are in the RFID transponders technology offered eg by Auto-ID-LabMIT The mentioned technology has become civilian development firstly in 1999However the first ideas on the modulation and magnetic survey of mini-antennas in theldquobugsrdquo belong still to Lev Thermin (research of electromagnetic and acoustic oscillationsin far ago year 1948) He is also the author of an exotic musical instrument ldquothereminvoxrdquocalled after his name and using the developed RFID principles
The next impulse to development was obtained from companies like SAP and Tele-foacutenica Further thanks to their ideas Cisco formulated the IoT creation conditions and
Fig 536 Fog computing primary concepts
186 5 Smart Grid Internet of Things and Fog Computing
Fig 537 Internet of things prognosis (Source Cisco)
basic requirements to IoT (Fig 537) It means amongst other things the urgent deploymentof IPv6 The usage of IPv6 with an available address space of 2128 addresses means thepossibility to address up to 3401036 active network devices or approximately 3001027
ldquothingsrdquo per user (figuratively each bacterium)Today the services provided by the Internet are also directly related to solving of the
problems of effective management of power and home control of embedded systems (smartfacilities intelligent homes) The efficient electricity consumption is considered in closeassociation with environmental and ecological problems which are regulated within theEuropean Union and the world community The regulatory basis is the internationallyratified Kyoto protocol an appendix to the United Nations Framework Convention onClimate Change and its follow-up meetings until 2015 in Paris According to the treatiessigned by 195 nations the global warmth must be restricted to C2 ıC over the pre-industrial levels Improving the energy efficiency of powerful computer servers and otherhousehold and industrial devices is achieved nowadays through the use of electricitydistribution networks and management solutions like smart grids [22]
Another important factor in the development of modern Internet services is the signifi-cant growth of the volumes of parallel computing combined with savings of computingresources Here the experts foresee firstly resources within the transition from cloud
52 From Internet of Services to Internet of Things Fog Computing 187
computing in some cases to the so-called Fog Computing which is associated withthe transfer of a large number of computing demands in the area of low-power homemicrocomputers Embedded processors -controllers and on-board computers have themain objective on effective management of consumer devices Fog computing and use ofmicrocomputers are directly related and can provide significant savings of energy Dueto the expansion of the concept of fog computing from the cloud computing paradigminto intelligent network nodes (so called Radio Network Edge) by network equipmentproducers such as Cisco a whole set of new applications and services was enabled Thefeatures of fog computing are as follows
bull node heterogeneitybull leading role of wireless accessbull low-latency location-awareness speed node re-activatingbull wide geographical distributionbull very big number of nodes and their mobility supported via IPv6bull priorised streaming and real-time applications
Fog computing offers the appropriate platforms for IoT-services clouds and smart gridsSuch networks provide automatic and automated execution of usual everyday routinesespecially domestic processes book reading listening to music home heating andairconditioning to make a cup of coffee to take medicine at regular terms to prepare andcook simple meals to water the flowers and garden and other activities with automationpotential It is because they insist on a combination of domestic hosts gadgets instrumentsand ldquothingsrdquo into a single heterogeneous network that will be served via low-energyldquogreenrdquo Internet protocols The use of traditional MAC and IPv4 addresses for data link andnetwork layers respectively can not identify an impressive number of deployed devicesTherefore there is no doubt that a gradual transition to IPv6 is required Started in 1990this transition seems to have accelerated since 2011 when many users switched fromtunneled IPv6 (6to4) to native connections leading to a 10-fold increase in adoptionjust three years later Still in 2015 the service provider Google reports that only about8 of requests to its services are delivered with IPv6 on a global level [7] The per-country statistics nevertheless show the different adoption speeds Belgium Switzerlandand Portugal each have more than 20 IPv6 traffic according to this statistic Cisco reportsother statistics however According to them these three countries each have more than45 IPv6 deployment [3]
Example 510 The Internet of Things (IoT) may be illustrated as follows Imagine acity or ordinary home diversity of smart gadgets (laptops smartphones and tablets)and multiple household appliances (TV alarm clocks coffee makers washing machinesrefrigerators microwave ovens automated window blinds) HVAC systems (boiler radia-tors air conditioning fan and ventilators) systems for garden irrigation security (lockscameras) and lighting systems (including solar panels) intelligent sensors (heat light
188 5 Smart Grid Internet of Things and Fog Computing
motion) and so on The warehouses delivery and logistic systems as well as publictransport and private cars have to be equipped in the long term with interfaces forWLAN3G Similarly the small ldquothingsrdquo (books compact discs DVDs medication inblisters fast food in vacuum packs soft drinks etc) can be equipped with low-costBluetooth interfaces RFID transponders and similar small-data links and then interact witheach other through further energy-efficient communications networks (infrared wirelessmobile power and low voltage networks)
521 Enabling Technologies for IoT
Dialectically enabling technologies help ldquoturning a quantity into a new qualityrdquo Thedemarcation of the categories of IoT IoS clouds and smart grids and the relatedones is given in Fig 538 In fact this demarcation is not quite clear nowadays Theconcepts are closely related and toothed due to their development histories The depictedconcepts are closely adjoined with modern methods and network technologies systemsand services given in ovals in the figure Since the use of cloud systems became wide-spread the ldquoInternet of Thingsrdquo has become a way of implementation and a platform for
Fig 538 Closely related demarcation through IoS cloud and fog computing IoT and smart grids
52 From Internet of Services to Internet of Things Fog Computing 189
fog computing with low-energy radion nodes That made an imperceptible architecturaltransformation from mixed-distributed decentralised powerful systems (voluminous andbig data processing clustering) to many small geographically distributed but logi-cally connected hosts gadgets appliances and ldquothingsrdquo into a single heterogeneousnetwork The number of devices (hosts gadgets) in todayrsquos Internet (of people) isby modern statistics about N 109 Thus the number of users corresponds to thepopulation of the earth Due to continued growth in the coming years the estimatednumber of devices will reach N gt 30 109 Therefore the qualitative change toIoT is possible faster than expected Accordingly to frequent estimations it should happenin 2020
The enabling technologies for IoT are manyfold Typically they are listed as follows
bull Mobile Networks (LTE 5G)bull GPS (Global Positioning System)bull Wi-Fi (Wireless Fidelity)bull WiMAX (Worldwide Interoperability for Microwave Access)bull Powerline Homeplugbull PoE (Power over Ethernet)bull KNX (Konnex) LON (Local Operating Network)bull Bluetooth IrDA (Infrared Data Association)bull WSN (ZigBee EnOcean)bull 6LoWPAN (IPv6 over low-power Wireless Personal Area Networks)bull RFID (Radio Frequency ID) NFC (Near Field Communication) QR (Quick
Response)bull Watermarks (as steganography applications)
In addition to the already discussed smart grid enablers the next fog computing technologyis combined via the use of energy-efficient protocols Being the interpenetration of IoTsmart grids and clouds fog computing is possible today eg on the basis of the energy-efficient and low-cost protocol 6LoWPAN that implements IPv6 over MAC protocols ofIEEE 802154 and PLC networks This protocol was standardised via IETF and is opento use via multiple vendors
Let us consider the most simple and price-efficient enabling technology In particularlet us put the focus on lowest-cost and simplest methods to IoT communication likethe RFID transponders (RFID tags) the Near Field Communication (NFC) tags and QR(Quick Response) labels Their function is to localise and connect the ldquothingsrdquo to Internetat large The RFID NFC and QR systems operate at short distances (10 cmndash10 m) and havetheir origins in logistics and warehousing Thanks to the energy efficiency of RFID andNFC the period of permanent service is rather long and is approximately 12ndash72 monthsAfterwards the batteries need to be replaced The extended capabilities for addressingthese free devices are provided by IPv6 which can support many IP nodes (devices) perone inhabitant of the world
190 5 Smart Grid Internet of Things and Fog Computing
RFID transponders The devices for reading of RFID (Radio Frequency ID) can beintegrated within modern smartphones as well as operate as standalone readers (RFIDreaders) just similar to multiple well-known card readers or bar code readers widely usedin trading and in the storage business The use of RFID transponders is regulated bythe International Telecommunication Union (ITU-T) and within the following assignedfrequency bands LW 125ndash134 kHz KW 1356 MHz UHF 865ndash869 MHz (in Europe)UHF 950 MHz (in USA and Asia) SHF 245 and 58 GHz Their constructions are veryvariative Usually RFID transponders (or RFID tags) are passive It means that in theircontstruction an excitation antenna is available (Fig 538) The other option for an activeRFID transponder is a more intelligent system with memory storage microcontroller andbattery Such systems have a shorter life expectancy but they can be programmed orconfigured to a suited smart grid or fog computing node Active transponders can thereforepublish data on their own without having to be polled The high frequency passive HFtransponders (RFID tags) use the well-known radar principle and through activation andmodulation of the magnetic field can carry out the survey code that RFID readers cancapture The antennas of HF transponders use also planar inductance coils with many turnsThe RFID transponders with sensorics are oriented to measure certain physical or chemicalparameters As a rule these are usually pressure acceleration expansion moisture orelectrical conductivity They need one of the RFID readers which possesses very differentconstructions handheld mobile fixed and combined with the bar code reader Commercialand logistics coding with codes in the 64 96 and 128 bit format is called EPC (ElectronicProduct Code) and is typically used in mass RFID transponders The deployment areas areas follows in municipal and warehousing on railways and airports in supermarkets andlibraries in logistics in animal tracking (eg dog tags) and in biometrical access controlsystems in particular an increasing number of international passports called e-passportswhich allow for crossing borders without border patrol staff
NFC and QR labels NFC (Near Field Communication) systems are supported by a widepalette of leading Operating System (OS) vendors for smartphones and tablets egWindows Phone 8 or higher Android 23 or higher as well as by API (Windows DeveloperProgram for IoT) There are the following two types of near-field communication whichare also visualised in Fig 539
bull without connection establishment within passive high-frequency transponders (HFRFID) based on the standards ISO 14443 and ISO 15693 this method is suitableexcept for applications working on sensitive data because on the phase of transponderactivation its antenna can be eavesdropped by third parties
bull connection-oriented (between two equal active transmitters Tx)
The QR labels (Quick Response) are designed for universal reading of small quantitiesof data They have become popular by encoding logical addresses in the form of URIsfor Internet applications in particular websites The operation principle for QR-reading
52 From Internet of Services to Internet of Things Fog Computing 191
Fig 539 Examples of fog computing with RFID
Fig 540 Operation principle for QR-reading mobile applications
mobile applications is depicted in Fig 540 First a camera sensor is directed at thedisplayed QR code Then a picture is taken and processed QR codes contain a certainamount of redundancy as well as positioning aids so that even under imperfect lightingand camera holding conditions the data will be retrieved In the final step the data isprocessed so that when it represents an URI a registered application is launched which inmany cases will be a web browser
192 5 Smart Grid Internet of Things and Fog Computing
Advanced Bluetooth v42 The Bluetooth (BT) Special Interest Group (SIG) wasfounded in 1998 by Ericsson IBM Intel Nokia and Toshiba The new specificationof BT released in 2014 and superseding previous BT versions including the onesstandardised as IEEE 802151 defines its advanced features towards smart grid IoT andfog computing use The specification differentiates between high performance and lowpower consumption use cases Its improvements are as follows
bull better privacy higher data ratebull IPv4IPv6 connectivitybull interoperability with 6LoWPANbull integration to an Internet Protocol Support Profile (IPSP)bull 25 faster transferbull 10 increased packet capacity (transmission errors power consumption is reduced)bull new deployment scenarios and further improvements for IoT
BTv42 uses additional data security techniques for BT connections eg the customersshould be informed in a shop about the proposals per beacons only if explicitly approvedIn BTv42 deployments IPSP uses IP based software infrastructures for managing of BTsmart devices BTv42 is ideal for IoT networked home environments required personalas well as large-room control Depending on the requirements there is the low-energyspecification (Bluetooth LE) the high-performance specification with enhanced data rate(Bluetooth EDR) and some devices even implement a dual mode which enables thecreation of adaptive applications
6LoWPAN This important enabling technology for smart grids and IoT acts simul-taneousely like a fog computing predecessor The acronym means ldquoIPv6 over LowPower Wireless Personal Area Networkrdquo There is a small genesis history of 6LoW-PAN Originally the company Jennic from Sheffield UK implemented the project6LoWPAN as equivalent to ZigBee equivalent The Jennic 6LoWPAN had the followingfeatures
bull standardised IETF IP networkingbull flexible topologiesbull SNAP API similar to SNMP
As such it is based on the IEEE 802154 WPAN standard and uses compressionmechanism to deliver IP packets efficiently over such links Most hardware supportsWPAN links in the 24 GHz band so that 16 channels and a data rate of 250 kbps areavailable The maximum transmission unit in such WPAN links is 127 bytes so that IPv6packets need to be fragmented into multiple WPAN packets
6LoWPAN networks can be set up in a point-to-point star and self-healing tree topol-ogy Typical cluster sizes are up to 100 nodes The protocol supports automatic staring
52 From Internet of Services to Internet of Things Fog Computing 193
clustering routing and healing and furthermore end-to-end message acknowledgementseven when routing in a mesh with multiple hops
For software development several APIs are provided The first is rather abstract andC-based for simple applications It gives access to the on-chip periphery and systemservices The second is called SNAP ndash Simple Network Access Protocol It works similarto SNMP and allows the IPv4IPv6-based data exchange between devices Likewise it isbased on a Management Information Base (MIB) for the remote control configuration anddiagnostics of 6LoWPAN networks
As an example a mesh network with Internet access by a 6LoWPAN gateway isgiven in Fig 541 6LoWPAN technology commonly resembles ZigBee although thereare important differences Most importantly as 6LoWPAN offers IP connections it iscompatible to protocols with other physical layers whereas ZigBee devices can onlycommunicate with other ZigBee devices The network stack implementation for theprotocol is therefore much smaller too which means more re-use and less sources forerrors Typical for 6LoWPAN is an IPv6 deployment for addressing a large numberof sensor nodes The large installation base of such devices led to the rise of the fogcomputing paradigm Nevertheless 6LoWPAN adoptions seems to be slower than ZigBeeadoption and other protocols such as ANT+ which is not even sharing the physical layerwith the other two are now gaining momentum which keeps the compatibility issue open
Fig 541 6LoWPAN Fog computing predecessor
194 5 Smart Grid Internet of Things and Fog Computing
Data security in IoT Some authors believe that the ldquoInternet of Thingsrdquo is a potentiallyldquodisruptiverdquo technology since it deals with the invisible widespread and ndash undesirablefor users ndash transformations to computing nodes (ie inter alia to small harmful ldquobugsrdquoor spies) of familiar and well-known ldquothingsrdquo like private cars walls of apartments andoffices electronic wares product packaging furnitures valuables conventional paperand more This transformation may violate the anonymity and private sphere of regularcitizens and even harm national data security Transformations to the IoT and fogcomputing are studied anyway thoroughly via leading political and power structures overthe world (EU Commission NSA in USA etc) The solution to this problem on datasecurity in IoT that appears just today is already possible through the use of relevantcryptoprotocols steganography and concealed routing within the IoT enabling wirelessnetworks and mobile networks Chapter VII is dedicated specially to the mentionedproblematics
522 Case Studies on IoT with On-Board Micro-controller Raspberry Pi
In the following paragraphs we offer case studies on the use of the on-board -controllerRaspberry Pi to realise low-energy systems for service delivery and fog computing
On-board -controllers of type Raspberry Pi Its compact size and low powerconsumption are the main priorities of the single-board computers such as AVR ArduinoIntel Edison and Raspberry Pi The models AA+BB+ 2B and Zero of Raspberry Pi areon-board -controllers that are oriented to mass usage for different areas of embeddedsystems IoT and smart grids Model B is shown in Fig 542
The Raspberry Pi node is normally coupled to a secured voltage block with a MicroUSBadapter The secured voltage block has the following work characteristics
bull Input Voltage ndash 90ndash264 V ACbull Voltage freuency ndash 47ndash63 Hzbull Output Voltage ndash 5 V DCbull Output current ndash up to 1200 mAbull Max power consumption ndash up to 6 Wbull Temperature ndash 0ndash40 ıCbull Dimensions ndash 64 485 255 mmbull Weight ndash 79 g
Table 55 printed below includes the comparison of the usual Raspberry Pi models A andB which offer a reasonable performance for running server applications and controllingconnected devices Compared to them the model Zero is much cheaper and smaller andwhile being faster than A and B offers less connectivity interfaces and is therefore moresuitable for software service delivery Model 2B is even faster despite lower tact due to its
52 From Internet of Services to Internet of Things Fog Computing 195
Fig 542 On-board computer Raspberry Pi model B (Source Oracle)
Table 55 Comparison of usual Raspberry Pi models A and B
Characteristics Model A Model B
Approximate price 25$ 35$
CPU 700 MHz ARM
GPU BroadCom VideoCore
Codecs H264 MPEG-2
SDRAM 256 MByte 512 MByte
Ports and interfaces ndash 2x USB30 1x SD 1xHDMI 1x Clinc TRS adapter6x GPIO
1x LAN Eth 10100 RJ45 2xUSB30 1x SD 1x HDMI 1xClinc TRS adapter 6x GPIO
Regular voltage cur-rent power
5 W 500 mA 25 W 5 W 700 mA 35 W
ARM Cortex-A7 CPU and equipped with more main memory but also more expensiveHence the choice of the right model depends on the use case and on the budget Theenergy supply can be also coupled via the microUSB cable Nominal voltage is 5 V thecurrent does not exceed the current 700 mA that is the regular power that it consumesis no more than 35 W Frequently instead of a hard disk the SD card is used as bootdrive The new SDHC standard allows capacities of up to 32 GByte The SD card has toretrieve the OS for the node as well as the necessary applications which can be installedfrom multiple freely-available ISO images for Raspberry Pi After image deployment the
196 5 Smart Grid Internet of Things and Fog Computing
re-configuration of the used services is possible depending on the use case The systemprovides a lot of adapters as well as ports (SD LAN USB HDMI GPIO Clinc)
Already announced is the Raspberry Pi 3 Model B which instead of requiring USBdongles has Bluetooth 41 (Low Energy) and WLAN adapters pre-installed [15]
The Raspberry Pi microcomputers are supported by many OS distributions Amongthem are adapted versions of existing systems such as Android Debian Ubuntu ArchLinux Gentoo and NetBSD but also dedicated distributions most prominently RaspbianRaspBMC (now OSMC) and Pidora Raspbian is based on Debian and tracks new modelsso that it is a good default choice One of the features of the system is a centralconfiguration file called configtxt to configure low-level parameters which wouldotherwise be configured in the BIOS Among them are display resolutions overclockingand USB power settings
The worldrsquos smallest PC and its applications The advanced Raspberry Pi acts alsoas the worldrsquos smallest PC in popular media as it symbolises the miniaturisation trendfrom clumsy PC hardware to embeddable micro-systems and nano-systems despite onlybeing one out of many single-board computers (Fig 543) This is especially the casefor the new Raspberry Pi 2 Model B which belongs to the type Mini-PC with 6 timesmore CPU performance in comparison to the conventional models The system can beequipped with the free-of-charge Windows 10 version as well as with the aforementionedOS distributions There are some constructive features of the Pi 2B
bull Broadcom SoC (System on Chip) BCM2836 which computes with the quad-core ARMCortex-A7 CPU
bull tact frequency reaches up to 900 MHz
Fig 543 AdvancedRaspberry Pi2 model B asmini-PC (Source chipde)
52 From Internet of Services to Internet of Things Fog Computing 197
bull larger RAM of 1 Gbytebull Support via Windows Developer Program for IoT in addition to free software OS
distributions
An application of Raspberry Pi is the deployment as low-energy home intelligent nodefor fog computing scenarios One of the most useful usage examples thus becomesthe energy-efficient service provisioning for XaaS (Everything as a Service) basedon these microcomputer units [14] The structure of these services can include interalia
bull sensor controllerbull home control systembull efficient small clusterbull private cloudbull file server and web server (Fig 544)
Fig 544 The examples of low-energy home intelligent node based on on-board -controllerRaspberry Pi
198 5 Smart Grid Internet of Things and Fog Computing
The microcomputer Raspberry Pi offers energy savings by consuming only up to 35 WTherefore with the use of Raspberry Pi it is possible to create energy-efficient XaaS asoutlined before But with such choices what is better Where are the avantages providedin more centralised often virtualised systems (clustering clouds) or small and moredecentralised ones (microcomputers piconets) To use big clusters or to start from theclouds multiple VMs from the hot reserve or the small on-board -nodes like RaspberryPi Arduino or Intel Edison with only small power consumption The discussed trade-offsherewith are as follows
bull reliability and QoSbull data security and privacy as well as access anonymitybull deployment effortbull energy consumptionbull operating expenses
There are no comprehensive answers to this question yet and it remains open today
Example 511 To create a media centre the Raspberry Pi 2 Model B is optimally suitedbecause it has a special unit that is responsible for the recognition of multiple codecs andformats XBMC Media Centre software can be recommended for this case XBMC MediaCentre is available across all OS options including Linux Mac OS X (Snow LeopardLeopard Tiger Apple TV) Apple iOS Microsoft Windows Android as well as pre-configured for Raspberry Pi The XBMC Media Centre uses diverse formats codecs andprotocols
bull graphic PNG JPEG BMP GIF ICO TIFF PCX etcbull audio MIDI AIFF WAVWAVE MP2 MP3 AAC AACplus AC3 DTS ALAC
AMR WMA etcbull video DivX Xvid BivX AVI MPEG-1 MPEG-2 H263 MPEG-4 MPEG-4 AVC
(H264) HuffYUV Indeo MJPEG RealVideo RMVB Sorenson WMV etcbull play lists PLS M3U WPLbull disk images CUE NRG IMG ISO BINbull network protocols IP IPv6 UPnP NFS SMBSAMBACIFS XBMSP DAAP HTTP
HTTPS FTP RTSP (RTSPU RTSPT) MMS (MMSU MMST) RTMP PodcastingTCP UDP SFTP RTP
bull media types CD DVD DVD-Video Video CD (VCDSVCDXVCD) Audio-CD(CDDA) Blu Rays USB Flash Drives HDD
bull meta-data APEv1 APEv2 ID3 (ID3v1 and ID3v2) ID666 Exif (GeoTagging)
One should also take into account that additional functional blocks affect the size of thedevice Therefore they should be realised in the form of individual hardware modules
52 From Internet of Services to Internet of Things Fog Computing 199
Fig 545 A media centre structure scheme based on Raspberry Pi
or anticipate making a special case which will be different from the standard pod forRaspberry Pi However making a separate connected device has significant advantages interms of practical use
In Fig 545 a media centre structure scheme based on Raspberry Pi is depicated Theconsidered media centre consists of the Pi node an HDMI monitor USB keyboard USBmouse infrared (IR) interface and speakers
The examined system based on Raspberry Pi is energy-efficient and offers the followingfeatures
bull video and audio players can access all files via FTP SFTP SSH and WebDAVbull multiple codecs that are retrieved from the SD card within a LAN or from the Internet
are supportedbull the IR control transceiver allows remote controlbull plugins for the integration with pupular online services are available
The new versions of XBMC are extended via an add-ons framework The extensions forXBMC Media Centre can be also implemented in the Python programming languagewhich makes this an easy task for IT-affine users The graphical user interface (GUI) forXBMC can be configured declaratively via WindowXML
Example 512 Let us examine the deployment of a web server on the basis of the on-board -controller Raspberry Pi The mobile and fixed network access to the service is
200 5 Smart Grid Internet of Things and Fog Computing
henceforth supported on PCs tablets and smartphones Taking into account the creation ofa cost- and energy-efficient host the use of a home DSL router from vendors like BelkinNetgear or Linksys among others is assumed to be possible What will the user haveto do as the next step The user needs to configure the system by using firmware (IPaddresses port 80 for the web server perhaps 8080 for additional services etc) and theninstall XAMPP for a linuxoid Raspberry Pi distribution The full package called XAMPPincludes inter alia
bull web server Apache with SSL supportbull MySQL Lite databasebull phpMyAdmin tool for the web-based administration of the databasebull PHP module for running server-side scriptsbull FTP client FileZilla for uploading content and scripts to the web serverbull ProFTPD daemon for offering an upload possibilitybull Perl module for more server-side scriptsbull servlet container Apache Tomcat with Java support for more complex server applica-
tionsbull mail server with POP3 and SMTP protocols and many more for additional services
The content management for the created web server as well as application support onthe Raspberry Pi micro-computer node is provided by using a Secure Shell client withthe associated protocol SSH With a client to Dynamic DNS (DDNS) the dynamicprovisioning and use of the domain name is enabled without evident registration by anInternet service provider Straight from the mentioned host the control of the creation anduse of the new server can be established eg in this manner laquomywebserverpublicdnsraquoFrom then on the web server and its content and applications are accessible to the world
Example 513 An example of an energy-efficient file server offering private cloud storagebased on the Raspberry Pi micro-computer unit is presented in Fig 546 Since the SD carddoes not have enough space and can not provide a stable long-term service with readingand writing oprations but rather requires a necessary external storage device a USB driveor network storage service can be controlled by the file server The system based on themicro-computer unit with the function file server includes the following elements
bull Raspberry Pi node with OS Raspbian or similar which is coupled to the Internet with aDSL router
bull an external USB drive with up to 5 TByte capacity (USB ndash SSDHDD such as SeagateBackup Plus) which is mounted as a hard disk drive with the tools provided by theoperating system
bull optionally more local or network drives to offer redundant storage with higher capacityandor higher availability
52 From Internet of Services to Internet of Things Fog Computing 201
Fig 546 Low energy file server based on a Raspberry Pi node
The file system of the storage device can be of any type considering that all clientsaccess the system through network protocols such as FTP SCP WebDAV SMB or CIFSTo set up the system software for such a flexible access it is necessary to use toolslike SSH Apache and Samba The Samba service is shipped by the majority of Linuxdistributions The main advantages of Samba are the free licencing simultaneous usageof different hosts within an IP networkLAN like Windows Unix and Linux with supportof file echange among them Under use of Samba an external storage device such as anUSB drive becomes ldquovisiblerdquo within the network de-facto like by the slogan laquoShare thedrive on your networkraquo
Clusters of Raspberry Pi A single on-board -controller is already quite capable Nowimagine a (Beowulf) cluster of these Prototyped at the Free University of Bolzano inSouth Tyrol Italy the Bobo with 40 nodes and the Bobino with 8 nodes (the model shownin Fig 547) combine cluster computing with tiny nodes [20] Apart from all nodes beingequal by running as workers some have been designed to assume special roles in order tokeep the system images lean and the system itself manageable The roles are (1) gateway(2) brain and (3) backup All nodes are internally connected by Ethernet The gatewayrsquostask is consequentially to connect the cluster to the outside world by Ethernet WLAN orEthernet-over-USB All internal processes are controlled by the brain node Finally the
202 5 Smart Grid Internet of Things and Fog Computing
Fig 547 Bobino a cluster of 8 Raspberry Pi nodes
backup node is queried to retrieve an unmodified image in case of accidental irreversiblemodifications during experiments
Such a system requires user-friendly node reservation grouping and monitoringfunctions The monitoring is essential because nodes may fail easily Imagine that eachnode has a mean time between failures (MTBF) of one million hours This means that theprobability of failure of any node in a two-year period is determined as follows [24]
pT D 1 e TMTBF D 1 e 2a
114y D 174 (510)
However the failure of the overall system depends on a serial MTBF in conjunctionwith the mean time to repair (MTTR) If the MTTR is too high the likelyhood of anothernode failing just when one is already under repair is quite high Therefore the followingholds
MTBFserial D1
1MTBF1
C 1MTBF2
C C MTTRMTBF1MTBF2
(511)
For the 8 nodes of Bobino and an assumed one-day repair this means that
MTBFserial D1
8MTBF C 24
8MTBF
D 9090909 h (512)
52 From Internet of Services to Internet of Things Fog Computing 203
In other words just about 1037 years Hence pT rises to 1753 For the 40 nodesof Bobo the values are correpondingly MTBRserial D 2463054 h or just about 281 yearsand pT D 509 meaning that a failure of the system is already more likely thanits continuous operation Parallel functionality with redundancy is therefore much bettersuited for such clusters
For the node reservation and grouping Bobo and Bobino ship with the MegaRPImiddleware which includes appropriate management web interfaces as well as user-oriented software including file managers on top of the standard Raspbian software
523 The Future Industry 40 Vision
Industry 40 platform Industry 40 (originating as Industrie 40 in Germany aroundthe year 2011) is a future strategic goal in the high-tech strategy of the German federalgovernment Its main driver is to advance the informatisation of the production processesThe goal is a smart factory characterised by adaptivity resource efficiency and ergonomicworking conditions as well as the integration of customers and business partners into thebusiness value chain The technological basis of industry 40 are cyber-physical systems(CPS) and the IoT cf Fig 548
Fig 548 Industry 40 as outlined by the German government program 2011 (Own representationbackground Google ldquoGreenrdquo Fabrics)
204 5 Smart Grid Internet of Things and Fog Computing
Fig 549 Industry 40 service visions (Own representation and photo)
Within Industry 40 information and communication technologies as well as automa-tion and production technologies become increasingly and more than ever dovetailed toeach other The political ambition is to defend and extend the traditional core of theGerman industry with its internationally outstanding positions as shown in Fig 549
524 Fog Computing
Fog computing as a concept means that the services data storages applications andcomputing (business logic) are shifted on the ldquonetwork edgerdquo ie closer to the usersonto interactive end devices or ambiental micro-factor devices The question which is tobe solved can be formulated as follows how close do they get partially or completelyThe other names for similar concepts are ldquoedge computingrdquo or ldquoeverything on the usersiterdquo The co-existence with cloud computing services is provided too The services areoffered in form of XaaS An example of a fog topology can be given as follows in Fig 550Despite a cloud typically operating as a central node the support of multiple intelligentfog nodes with the shifted functionality is foreseen
52 From Internet of Services to Internet of Things Fog Computing 205
Fig 550 Topology for fogcomputing
Fig 551 Cloud and fog computing common architecture
A common architecture for combined use of Cloud and Fog computing is depicted inFig 551 The architecture includes the following three hierarchical planes
1 Plane 1 The clouds and data centres which build an IoS with typical access via webservice protocols
206 5 Smart Grid Internet of Things and Fog Computing
Table 56 Fog advantages Requirements Advantage
Low latency Less hops
High data mobility Data locality and local caches
Less limited data rate On-site processing
Reliability and robustness Fast failover
Rich storage with metadata Location awareness
2 Plane 2 The fog nodes which are involved to the virtual environments for datapreprocessing functionality migration and load balancing with the clouds (refer plane1)
3 Plane 3 The users with end devices which build an IoT and are placed on the edge ofthe fog infrastructure
Such kind of the distributed architectures for combined use of cloud and fog computingoffers several clear advantages Table 56 summarises them specifically for requirementson cloud and network storage The main requirements on fog computing on a technicallevel are as follows
bull IPv6 deployment to reach millions of serving devicesbull growth of provided security in particular deployment of firewalls and intrusion
detectionbull authenticity of coupled devices must be guaranteed everywhere in the combined
structures (users + fog + clouds)bull encryption and digital signature has to be guaranteed via robust combinations of
AES+RSA+PKI
Concrete technical platforms for fog computing are rare They remain mostly a vaguetechnical concept to be fully realised within the next years Still a few preliminaryarchitectures exist One such implementation platform to cloud and fog computinginteroperability is offered in [46] and shown in Fig 552 Suitable network option for theplatform are ZigBee EnOcean 6LoWPAN coupled with cheap microcontrollers
53 Conclusions
The chapter discussed the architectural transformations of modern networks and theirmobile services and applications in the framework of development of upcoming networktechnologies like ldquoSmart Gridrdquo (as an intelligent network for services as electricity andenergy-efficient information services) as well as ldquoInternet of Thingsrdquo IoT (providing radio-communication of multiple milliards of low-power IPv6 devices at near distance) withtheir methods of implementation in the form of ldquoFog Computingrdquo
53 Conclusions 207
Fig
55
2Fo
gco
mpu
ting
plat
form
and
appl
icat
ions
with
clou
dco
nnec
tivity
(Fro
m[4
6])
208 5 Smart Grid Internet of Things and Fog Computing
In some developed countries an integrated intelligent network on the sample of theconventional Internet is rapidly created (a network with open mesh platforms for energyservices) The network possesses the ability to use standardised software interfacesas well mobile applications with several offered web services and among them cloudservices Thanks to the standardisation of smart grid (accordingly to the intentionsof the organisations like NIST IEEE VDE CENELEC etc) software and hardware-independent access and communication between the components are although not yetguaranteed quite likely Nevertheless some devices only communicate with proprietaryprotocols to send data to services determined by their vendors which severely restricts theubiquitous connectivity visions
The standardisation of the structure of the open networks towards smart grids is todayone of the development priorities as for energy and telecommunications industry in boththe USA and Europe The combined services of such networks will find in the near future(about 2020ndash2030) an opportunity to attract a stable increasing number of stakeholdersand users Nowadays there is the opportunity to create a large range of its own ldquosmartapplicationsrdquo and ldquosmart servicesrdquo within the smart grids
Thus to the development of such integrated electric power networks and telecommu-nications both will soon be given a necessary impulse The smart power grid services (ieelectricity) will be freely delivered disposed to the market and freely traded there frommultiple perspectives purchase sale exchange credit providers and resellers The effectwill be analog to todayrsquos ongoing revolution of smartphones and tablets on the mobilecommunication market that has arisen as a result for instance of deployment of alreadyfamiliar and contemporary concepts like the application directory App Store (Apple) oropen source OS Google Android
It is expected that the integration technologies and models for electrical networksand telecommunications discussed in this work will lead to a reduction of the overallconsumption of conventional energy sources CO2 footprint under the Kyoto protocolto further decentralisation of the supplier networks (based on the principle of Internetconstruction) Smart grids have to increase in the middle-term the energy efficiency underuse of alternative and renewable sources like wind solar and EM-smog They will inspireoptimisation techniques for network management and service billing (smart metering)for the integrated networks for power supply systems and telecommunication both byincreasing of its safety security and QoS
The decisive importance of smart grids and the IoT is the use of wireless networks likePowerline ZigBee EnOcean and 6LoWPAN and components with established servicesfor measurement automation and parameters control (so-called smart metering) whichconverts the parameters of the environment and climate to digital form
Now that the worldrsquos leading IT companies are engaged in the implementation of smartgrids and cloud computing for example Google with Nest and the Compute Platform oneof the major problems remains the studies of the opportunities and challenges of alternativeenergy sources in order to create environmentally friendly technologies and to improve theclimate on the planet
References 209
References
1 Bundesministerium fuumlr Wirtschaft und Energie online httpbmwide2 CISCO Grid Operation Solutions online httpwwwciscocom3 Cisco 6lab - The place to monitor IPv6 adoption online http6labciscocomstats 20154 Comiteacute Europeacuteen de Normalisation Eacutelectrotechnique online httpwwwcenceneleceu5 Energieinformationsnetze und -Systeme Bestandsaufnahme und Entwicklungstendenzen 2010
128 p in German ITGVDE6 EU Commission Expert group on the security and resilience of communication networks and
information systems for smart grids online httpwwwsmartgridseu7 Google IPv6 Statistics online httpwwwgooglechipv6statisticshtml 20158 Ibh it-service gmbh online httpswwwibhde 20159 IEEE Smart Grid Conceptual Model online httpsmartgridieeeorg
10 Kiwigrid Smart Grid Management Platform online httpwwwkiwigridcomenproducts-solutionshtml 2016
11 NIST Framework and Roadmap for Smart Grid Interoperability Standards Rel 20 TechnicalReport 1108R2 National Institute of Standards and Technology USA February 2012
12 OECD Digital Economy Outlook online httpsdxdoiorg1017872F888933225312 May2015
13 Projects of the Chair of Computer Networks of TUD online httpwwwrninftu-dresdende14 Raspberry Pi Projects online httpelinuxorgRPi_Projects 201615 Raspberry Pi Trading Ltd Raspberry Pi 3 Model B - Single Board Computer online https
fccidio2ABCB-RPI32 201616 Siemens AG online ttpwwwsiemenscom17 Smartgridgov online httpswwwsmartgridgov 201518 Technisch-wissenschaftlicher Verband der Elektrotechnik und Elektronik online httpwww
vdecom19 Uptime Institute Reports 2011ndash2014 online httpsuptimeinstitutecom20 Pekka Abrahamsson Sven Helmer Nattakarn Phaphoom Lorenzo Nicolodi Nick Preda
Lorenzo Miori Matteo Angriman Juha Rikkilauml Xiaofeng Wang Karim Hamily and SaraBugoloni Affordable and Energy-Efficient Cloud Computing Clusters The Bolzano RaspberryPi Cloud Cluster Experiment In UsiNg and building ClOud Testbeds (UNICO) workshop at the5th IEEE International Conference on Cloud Computing Technology and Science (CloudCom)volume 2 pages 170ndash175 December 2013 Bristol United Kingdom
21 Joumlrg Benze Smart Grid Normung und Standardisierung 2012 FH Salzburg IKT Forum22 Brussels EU-CEN-CENELEC-ETSI SG Coordination Group Smart Grid Reference Architec-
ture Technical Report M490 CENELEC November 2012 p 10723 S Guy S Marvin W Medd and T Moss Urban Infrastructure in Transition Networks
Buildings Plans EarthscanRoutledge London 2012 240 p24 Thomas J Harrison and Thomas J Pierce System integrity in small real-time computer systems
In Proceedings of the national computer conference and exposition (AFIPS) June 197325 Horst Kuchling Taschenbuch der Physik Hanser Verlag 2014 21st edition 711 p in German26 R Lehnert Smart Grid Communications In Proceedings of IEEE ELNANO Conference Kiev
Ukraine April 201327 Andriy Luntovskyy Integration Concepts for Computer-Aided Design Tools for Wired and
Wireless Local-Area Networks Shaker Verlag Aachen 200828 A Luntovskyy Distributed applications technologies DUIKT Publisher 2010 474 p
Monograph in Ukrainian
210 5 Smart Grid Internet of Things and Fog Computing
29 Andriy Luntovskyy Dietbert Guumltter and Igor Melnyk Planung und Optimierung von Rechner-netzen Methoden Modelle Tools fuumlr Entwurf Diagnose und Management im Lebenszyklus vondrahtgebundenen und drahtlosen Rechnernetzen SpringerVieweg + Teubner Verlag Wiesbaden2011 411 p in German
30 A Luntovskyy M Klymash and A Semenko Distributed services for telecommunicationnetworks Ubiquitous computing and cloud technologies Lvivska Politechnika Lviv Ukraine2012 368 p Monograph in Ukrainian
31 Andriy Luntovskyy Josef Spillner and Volodymyr Vasyutynskyy Energy-EfficientaNetworkServices as SmartaGridaIssue In Soft Computing in Computer and Information Science Advances in Intelligent Systems and Computing volume 342 pages 293ndash308 SpringerInternational Publishing Switzerland March 2015
32 Harald Lutz and Ulrich Terrahe Future Thinking Kongress Das Rechenzentrum der Zukunft33 V Melnyk Modeling of the temperature modes for the cathodes of high voltage glow discharge
based on heat balance equation Bulletin of Kherson National University of Technology Issue 3(39) 2010
34 Igor Melnyk and Andriy Luntovskyy bdquoGreen Computingldquo and the Simplified Waste HeatTransport Models In 20th International Conference on Advanced Computer Systems (ACS)2016
35 J Momoh Smart Grid Fundamentals of Design and Analysis John Wiley amp Sons NY 2012216 p
36 Bryan Nicholson Becky Harrison and Lee Cogan The future of the grid ndash evolving tomeet americarsquos needs online httpswwwsmartgridgovfilesNortheast-Region-Workshop-Summary-Finalpdf May 2014
37 J Ploennigs V Vasyutynskyy and K Kabitzsch Comparative Study of Energy-EfficientSampling Approaches for Wireless Control Networks IEEE Transactions of IndustrialInformatics (TIT) 6(3)416ndash424 August 2010
38 Alexander Schill and Thomas Springer Verteilte Systeme - Grundlagen und BasistechnologienSpringer-Verlag second edition 2012 433 p in German
39 Rene Marcel Schretzmann Jens Struckmeier and Christof Fetzer CloudampHeat Technologiesonline httpswwwcloudandheatcom 20112014
40 Matt Stansberry 2014 Data Center Industry Survey online httpsjournaluptimeinstitutecom2014-data-center-industry-survey 2015
41 L Stobbe M Proske H Zedel R Hintemann J Clausen and S Beucker Entwicklung desIKT-bedingten Strombedarfs in Deutschland Studie im Auftrag des Bundesministeriums fuumlrWirtschaft und Energie Fraunhofer IZM and Borderstep Institute 2015
42 Andrew S Tanenbaum and David J Wetherall Computernetzwerke Pearson Studium fifthedition 2012 1040 p in German
43 S Tugay Mathematic modeling of the physical processes on the surface of the cooled cathodesin the electron sources of high voltage glow discharge Electron Simulation Vol 34 No 62012
44 Katherine Tweed China Pushes Past US in Smart Grid Spending IEEE Spectrum EnergywiseBlog February 2014
45 V Vasyutynskyy and K Kabitzsch Event-based Control Overview and Generic Model In IEEEInternational Workshop on Factory Communication Systems (WFCS) pages 271ndash279 NancyFrance May 2010
46 Shanhe Yi Zijiang Hao Zhengrui Qin and Qun Li Fog Computing Platform and ApplicationsDept of Computer Science College of William and Mary 2015
6Future Mobile Communication From 4G To 5G 5GEnabling Techniques
Keywords
Mobile cellular and satellite radio networks bull 4G bull 5G bull Enabling technologiesand inter-operability bull IoT bull QoE bull Future standard IMT 2020 bull Distributed InputDistributed Output (DIDO)
61 Conventional Techniques
Conventional telecommunication technologies integrate mobile cellular and satellite radionetworks and are typically divided into four generations by most of the literature (Fig 61)The peak data rates are depicted below within the figure The next generation 5G will bedeployed in the mid-term although most likely after 2020 due to the high developmentcost and the ongoing amortisation of the predecessor 4G [17 18]
The generations (shorthand G) started with 1G and 2GGlobal System for MobileCommunications (GSM) with some obsolete extensions (as a basis) Soon afterwards3GUniveral Mobile Telecommunications System (UMTS) and the accelerator HighSpeed Download Packet Access (HSDPA) (sometimes referred to as 35G) was rolledout and is nowadays practically deployed world-wide 4GLong-Term Evolution (LTE)has then been introduced as current standard with a recent upgrade to LTE Advanced Inthe meantime research activities concentrate on the coming-soon 5G introduction withina future standard International Mobile Telecommunications (IMT) 2020 Cellular radionetworks enable division of geographic areas into radio cells with specific frequencybands The current 3G4G architecture of mobile communication including WPANWireless Local Area Network (WLAN) WiMAX etc is extended with a hierarchicalcell structure down to picocells and femtocells [15] (Table 61) Cells refer to the signaltransmission radius around an antenna The larger the cell the less the number of installed
copy Springer Fachmedien Wiesbaden GmbH 2017A Luntovskyy J Spillner Architectural Transformations in Network Services andDistributed Systems DOI 101007978-3-658-14842-3_6
211
212 6 Future Mobile Communication From 4G To 5G 5G Enabling Techniques
Fig 61 Generations of mobile communication
Table 61 Hierarchical cell structure for mobile communication
Type DistanceData rate(MBits)
Mobility(kmh) Deployment in 3G and 4G
Giga Cell 100 km 0144
1013 kms or4700
Transnational providers satellites
Macro Cell 10 km 0384 2 500 National providers
Micro Cell 1000 m 0384 72 120 Campus city districts metropolitanareas
Pico Cell 100 m 72100 10 Hotspots ndash railway stations cafesairports hotels
Femto Cell 10 m 28 10 Residential gateways
antennas needs to be but at the same time larger cells would mean a higher number ofrecipients causing issues with signal strength and connection management Femtocellsare the smallest cell size in use They accomodate a low number of connections (up to 16)mostly in residential settings and hence are comparable with WLAN
Example 61 According to Swisscom a Swiss telecommunications networks operatorthe needs-driven bandwidth evolution happened in the following way [5] Free voice callsover the Internet summed up to 750 billion minutes in 2013 and will increase to 1700billion in 2018 In 1993 voice transmission over the Internet was not yet feasible as the2G (GSM) bandwidth was 02 MBits In the 3G time introduces with UMTS in 2001 thebandwidth increased to 039 MBits then in 2008 with HSPA to 72 and two years later
61 Conventional Techniques 213
with HSPA+ even to 42 MBits The 4G (LTE) time started in 2011 with 150 and peakedthrough LTE Advanced in 2014 with even 450 MBits
Due to their current technology both LTE networks and satellite radio systems will bepresented in greater detail on the next pages
611 LTE Networks
The advantages of 4G or Long Term Evolution are nowadays as follows
bull compatibility to UMTSHSDPA and moderate to higher data rates as a rule up to300 MBits downlink and 75 Mbits uplink
bull LTE spectral efficiency 13 BitsHz vs only 02 by 3Gbull deployment of advanced techniques on modulation and antennas like Orthogonal
Frequency-Division Multiplexing (OFDM) and Multiple Input ndash Multiple Output(MIMO) antennas
bull flexible channel bandwidths (from 14 MHz up to 20 MHz)bull very low latency of less than 5 msbull deployment of unified IP Multimedia Subsystem (IMS) platform
The IMS uses the Session Initiation Protocol (SIP) specified in Requests for Comments(RFC) 3261 to offer telephony services as a combination of conventional switched-circuit networks and Internet Protocol (IP) networks The system architecture of LTE C
IMS is given in Fig 62 The basic components of LTE architecture are as follows
bull SGSN ndash Serving GPRS Support Node (GPRS)bull SAE ndash 3GPP System Architecture Evolutionbull GERAN ndash GSM EDGE Radio Access Network (EDGE)bull UTRAN ndash UMTS Terrestrial Radio Access Network (UMTS)bull IMS ndash IP Multimedia Subsystembull PSS ndash Packet-switched Streaming Servicebull PCRF ndash Policy and Charging Rules Functionbull EPS ndash Evolved Packet Systembull EPC ndash Evolved Packet Corebull HSS ndash Home Subscriber Serverbull MME ndash Mobility Management Entitybull IASA ndash Inter-Access System Anchorbull UPE ndash User Plane Entity
The current performance for LTE downlink in several countries is compared inTable 62
214 6 Future Mobile Communication From 4G To 5G 5G Enabling Techniques
Fig 62 4GLTE architecture
Table 62 4G downlinkperformance
International 75 MBits
Korea 186 MBits
USA 65 MBits
The system is based on GPRS EDGE UMTS technologies (GERAN UTRAN SAE)and is completely packet-oriented The IMS platform enables Voice over IP (VoIP) withsupport of conventional protocols (cp Fig 63) as well as data services on the base of SIPand other standardised protocols
Within IMS different planes or layers are defined The first one is the user plane orgateway which connects the system to an IP uplink The second one is the control planeor gateway control Through this plane caller identification and billing information isexchanged The third one is call control or session control The fourth one is the servicesfunction plane Among other tasks it contains functions to check the connection qualityfor emergency calls the connection to messaging services (SMS) and to connect prepaidcallers to the system The Diameter protocol (RFC 6733) is used within IMS to perform theauthentication authorisation and accounting of communication partners It succeeds thepreviously used Radius protocol which is however still in use in WLAN roaming networks
61 Conventional Techniques 215
Fig 63 General architecture for conventional protocols for VoIP and multimedia
and other constellations The simplified layered IMS architecture with the planes (a) andservice components (b) including classical fixed networks is depicted in Fig 64
612 Satellite-Based Radio Systems
The 4G architecture is also augmented with satellite-based radio systems (Fig 65) Thegeneral features of satellite-based radio systems are as follows
bull large latencybull large bandwidthbull many channelsbull time division algorithms
The radio systems are often only usable with a large latency about 024 s with GEOsThis severely impacts real-time communication but the remaining features still makeit suitable for other communication requirements The satellites typically offer separateuplink and downlink bands either 46 GHz or 1214 GHz These huge bandwidths are
216 6 Future Mobile Communication From 4G To 5G 5G Enabling Techniques
Fig 64 (a) Planes (b) Service components AS ndash Application Server SCIM ndash Service CapabilityInteraction Manager MRFC ndash Multimedia Resource Function Controller MRFP ndash MultimediaResource Function Processor MRF ndash Media Resource Function CSCF ndash Call Session ControlFunction BGCF ndash Breakout Gateway Control Function MGCF ndash Media Gateway Control FunctionMGW ndash Media Gateway HSS ndash Home Subscription Server HLR ndash 2G Home Location RegisterSimplified IMS architecture
61 Conventional Techniques 217
Fig 65 Satellite-based radio systems (Based on rninftu-dresdende)
oriented at eg each 500 MHz and each 50 Mbits thus enabling broadband commu-nication As a general observation the channel structure consists of 800 digital voicechannels with 64 kbits (800 64 D 50000 kbits data channels) Their allocationhappens for short time periods to individual channels through time division multiplexingon-demand
Satellite-based radio systems architecture includes the following components
bull GGW ndash Gateway Ground Stationsbull Footprint as a general covering or service areabull Spotbeams which are placed by each satellite as service areabull ISL ndash Inter-Satellite Linksbull MUL ndash Mobile User Linksbull GWL ndash Gateway Linksbull the IP backbone which is implemented via convenient DSL MPLSATM as well as
regional-specific technologies (eg HSDPA)
The motion of the satellite transponders can be described with good proximity via theplanetary motion theory basically elaborated by Johannes Kepler Galileo Galilei andNicolaus Copernicus Therefore we can use the following formulae
Angular frequency
D 2 f T D1
fD
2
(61)
218 6 Future Mobile Communication From 4G To 5G 5G Enabling Techniques
Gravitation on Earth
FG DMm
R2(62)
By Newton
FG D gm (63)
Therefore
g DyM
R2(64)
Transformed because g and R are known constants
M D gR2I FGr DMm
r2D gm
R
r
2
(65)
Furthermore it is important to demarcate the satellite height (h) from the distance toEarthrsquos middle point (r)
r D R C h (66)
The satellites describe an elliptical or circular orbit around the Earth The height h (thedistance r from the Earth center) remains constant because
FG D mg
R
r
2
D mr2 D FZ (67)
whereFG ndash Attraction of earth FZ ndash Centrifugal force m ndash Mass of the satellite R ndash Earth radius
6370 km r ndash Distance to earth middle point g ndash Acceleration of gravity g = 981 m=s2 ndashAngular frequency D 2 f T D 1=f D 2= f ndash Rotational frequency of the satelliteM ndash Mass of earth ndash Keplerrsquos constant
As a brief conclusion herewith is Keplerrsquos Law
a DgR2
42D const a D
r3
T2(68)
The formulae 7 C 8 solved for r offers (9)
r D3
sgR2
2 f 2(69)
61 Conventional Techniques 219
Where the distance from a satellite to the earthrsquos surface depends only on its orbitalperiod In the special case with T D 24 h with synchronous distance and specificallyh D 35786 km it means (example visualised in Fig 66)
r D 6370 km C 35786 km D 42156 km (610)
The classes of satellite-based radio systems are called GEO MEO LEO and HEO andthey are depicted in Fig 67
The comparison oft the satellite-based systems is given in Table 63 and Table 64The most important data for the current and historical types of satellite-based systems are
Fig 66 Explaining thecontext of r and T in KeplerrsquosLaw (Based on rninftu-dresdende)
Fig 67 LEO ndash Low Earth Orbit MEO ndash Medium Earth Orbit HEO ndash Highly-Elliptical OrbitGEO ndash Geostationary Earth Orbit Satellite system classes GEO MEO LEO and HEO (Based onrninftu-dresdende)
220 6 Future Mobile Communication From 4G To 5G 5G Enabling Techniques
Table 63 Examples of radio SAT
SATsystemtype Class Orbit h
Number ofSAT F-Band DR max Services
Orbcomm LEOoriginallycommer-cial2000
775ndash800 km
27 smallsatellitesm=45 kg2G ndash since2014further 18
VHF band137ndash150 MHz
48ndash576 kbits
EmailsTelephony
Inmarsat GEO since1979commercial
35786 km 5ndash11 fivegenera-tions
ndash 492 kbits Navigation TVInternet links Seaemergencycommunicationservices AirTraffic ControlGPS EGNOS
Globalstar LEO1991ndash1994
1400 km 48+4 ndash 144 kbitsviachannelbundling
Telephony datatransfer
ICO RTT MEO1998ndash2000
10390 km 10+2 ndash ndash Telephony datatransfer
Teledesic LEO1997ndash2002
700 km 288m=120 kg
286ndash291 GHz
100 MbitsUL720 MbitsDL
TelephonyInternet links
Iridium LEO1997ndash1998
780 km 66 (+6) ndash 24 48 kbits
Telephony datatransfer
summarised regarding to class services and deployment area transponder multiplicitylicenced frequency band orbit height and circulation period data rate transmitting powerlatency and operation durability
The GEO SAT systems (Fig 68) operate on constant distance to the Earth and possessa relatively high latency
D2 h
cD
2 35786 km
300 000 kms
D 0239 s (611)
The non-stationary LEO SAT systems are characterised as follows
bull distance h from the Earth of ca 300ndash1800 kmbull shorter signal propagation times (5ndash10 ms)bull lower transmission power of mobile stations sufficiently
61 Conventional Techniques 221
Table 64 Comparison of radio SAT
Satellitesystems GEO MEO LEO
Distance km h D 35786 kmr D 42156 km
r-R D 6000ndash12000 kmrespectively20200
r-R = 300ndash1800 km
Periode T 24 h 6ndash12 h 90ndash120 min
Latency t 025 s 70ndash80 ms 10 ms
Transmittingpower W
10 5 1
Deployment Multiplicity on systemsca 2000 Sputnik(1957) Intelsat 1ndash3(1965 1967 1969)Marisat (1976)Inmarsat-A (1982)Inmarsat-C (1988) etc
ICO 10+2 Iridium 66+6 Globalstar48+4 144 kBits Teledesic(2003) 288 2ndash64 MBitsOrbcomm 35
Bitrate kBits 01ndash1 10 1ndash64000
Average lifetime years
15 10 5ndash8
bull however more satellites required (gt50) frequent handover between satellites (aboutevery 10 min)
bull short lifetime of the satellite due to atmospheric friction (only 5ndash8 year)bull examples Iridium Teledesic Globalstar ISS (Fig 69)
MEO SAT systems are operated generally in the distance about 10000 km and have alower required number of satellites (about 12) They are characterised with slow motionno frequent handover between satellites is necessary The period is T D 6 h MEO providesan average life time under 10 years The problems of using MEO are as follows
bull propagation time 70 to 80 msbull higher transmission power necessarybull special antennas required
As an import and well-known MEO system class the navigation satellites have to bediscussed The examples are as of early 2016
bull GPS (USA) h D 20200 km T 12 h 32 satellitesbull GLONASS (RF) h D 19100 km T 11 h 15 min 28 satellitesbull GALILEO (EU) h D 23222 km T 14 h 30 satellites
222 6 Future Mobile Communication From 4G To 5G 5G Enabling Techniques
Fig 68 GEO SAT systems
62 A New Generation of Mobile Communication
One of the most popular definitions for 5G as a new generation of mobile communicationis as follows ldquoIn evolutionary view it will be capable to support wireless WWW allowinghighly flexible dynamic ad-hoc wireless networks in revolutionary view this intelligenttechnology is capable of interconnecting the entire world without limitsrdquo [7] While thisdefinition is very broad it emphasises new requirements and motivates us to take anotherlook at the mobile communication generations
A comparison of the existing mobile network generations is given via Table 65The network specialists from Deutsche Telekom NTT DoCoMo Amtel Samsung
Telefonica Vodafone Ericsson and other telecommunications operators [14] generateurgently their visions and technical requirements for future generation mobile commu-nication as well as the new standard 5GIMT 2020 The research on 5G technology beganin 2012 in France with achieving data rates over 4 GBits
In 2013 in Japan a new step towards 5G was made the equipment of the companyNTT DoCoMo has shown the ability to transfer data from the user with a data rateof up to 10 GBits (uplink) at a frequency F D 11 GHz on the 400 MHz bandwidthData was carried on the vehicle at a speed of 9 kmh In October 2014 the companySamsung Electronics has made a new recent record-breaking experiment with a datarate of 12 GBits at a vehicle speed of 100 kmh and even a data rate of 75 GBits in
62 A New Generation of Mobile Communication 223
Fig 69 (a) ISS as special LEO (b) Humanityrsquos first space flight on 1241961 durabil-ity D 108 min height h = ca 400 km (LEO) LEO SAT systems (Sources reflektioninfoNASA)
Table 65 Mobile generation comparison (Source wwwelektronik-compendiumde)
Generation Radio technology Transfer type Data rate
1G AMPS Analog circuit switching obsolete ndash
2G GSM Digital circuit switching 96 kbits
25G HSCSD Digital circuit switching 576 kbits
GPRS Digital packet switching 115 kbits
275G EDGE Digital packet switching 236 kbits
3G UMTSUTRAFDD
Digital mostly packet switching 384 kBits
UMTSUTRATDD
Digital mostly packet switching 2 Mbits
35G HSPA (HSDPAHSUPA)
Digital packet switching 144 Mbits
39G LTE Digital packet switching 150 Mbits
4G LTE Advanced Digital packet switching actual stan-dard
1 Gbits
5G IMT2020 Digital packet switching 10 100 Gbits
stationary conditions at a frequency of 28 GHz But the use of such higher frequenciesby F gt 5 GHz (in the mm-band) is rather problematic due to large attenuation in denseurban areas without increasing the transmission power On the other hand low-frequencytransmission is not always possible necessary licenses and (inter-)national regulationsare obstacles Therefore other new methods and international voting and conventions arerequired Samsungrsquos mm-wave testbeds set up in October 2014 have shown (as visualisedin Fig 610) the following results [14]
224 6 Future Mobile Communication From 4G To 5G 5G Enabling Techniques
Fig 610 Advanced communication technologies for high speed mobility (Source SamsungElectronics)
bull data rate approximately 2 GBits by velocity of 110 kmh was the worldrsquos first 5G datatransmission at highway speeds
bull record-breaking 12 GBits data transmission was reached at over 100 kmhbull in stationary conditions under use of F D 28 GHz spectrum the data rate 75 GBits was
obtained
621 Visions and Requirements
The official 5G start is planned to happen only in the year 2020 The status nowadays(architecture depicted in Fig 611) is as follows
bull research on advanced antenna techniques interference minimisation and further devel-opment of enabling technologies towards 5G (see next sections)
bull world-wide activities and tests among them Ishigaki (NTTDoCoMo) Seoul (Sam-sung) Stockholm (Ericsson) Dresden (Vodafone Chair 5glabde) London (KingrsquosRoyal College) Lund University (Sweden) BeijingShenzhen (China) and others (seeFig 611)
Requirements for the 5th generation The main 5G requirements are as follows
bull use of existing 4G infrastructure with augmentation via flexible WLAN-conformcommunication everywhere under international voting and conventions
62 A New Generation of Mobile Communication 225
Fig 611 4G with SAE ndash 3GPP System Architecture Evolution GERAN ndash GSM EDGE RadioAccess Network (EDGE) UTRAN ndash UMTS Terrestrial Radio Access Network (UMTS) IMS ndash IPMultimedia Subsystem SDN RAT ndash Radio Access Technology (Handover) DIDO for Multiuser-Wireless MIMO the systems with multiple Tx Rx antennas The 5G basic architecture
bull medium term obtaining of data rate D 10 GBits this rate corresponds to up-to-dateneeds to multi-media content download
bull tiny latencies real time inter-operability services without human interventionbull wide use of available frequency bands mm-Band with F D 30 up to 300 GHz (partially
and questionable)bull inter-operability with further mobile and wireless radio networks
The advanced antenna technique MIMO was already deployed in diverse network tech-nologies like WiMAX 80216adem WLAN 80211nacad LTE and others MIMOantennas allows nowadays communication with NTx D 16 transmitting and NRx D 16
receiving antennas Thus also a downlink with a data rate of DR D 10 GBits andabove is possible This DR D 10 GBits is about one hundred times fater in contrast toDR D 100 MBits the current status of peak data rate of LTE For the standard IMT20205G the wide use of 3D arrays for multiple input and multiple output channels (MIMOup to 16 16 16) is foreseen [3] The related data rates and mobility for mobile usersin the mobile communication systems of 3G 4G and 5G is depicted in Fig 612 Theprovided data rate will be increased more than 5000 times The peak data rate will thusreach 50 Gbits The data rate must be increased 10 up to 50 times in comparison to theones offered by LTE and LTE Advanced The prognosis is as follows in 2020 up to 50
226 6 Future Mobile Communication From 4G To 5G 5G Enabling Techniques
Fig 612 From 3G to 5G Datarates to mobility (By Samsung Electronics)
milliards devices will be IPv6-driven partially with 5G So for instance the priority of5G directions for companies in the telecom manufacturing area for instance Ericsson areas follows
bull digital economy remote machine controlbull smart gridsmart meteringbull Internet touch technologies smart citiesbull and IoT (Internet of Things)
The ongoing 5G forums for the advancement of specifications and testbeds for futuretelecommunication protocols are as follows
bull 5G PPP (5G Infrastructure Public-Private Partnership)bull METIS (Mobile and wireless communications Enablers for Twenty-twenty (2020)
Information Society)
The research laboratory 5glabTU Dresden There are multiple 5G activities inseveral universities and research laboratories in addition to the commercial researchactivities by telecommunications equipment manufacturers One such laboratory has beenestablished in Dresden Germany At Dresden University of Technology a modern 5GLaboratory at the Vodafone Chair for Mobile Communications Systems has openedto advance the data rates coverage connection stability and other aspects of mobileconnections [113] The researchers can evaluate and test a broad spectrum of 5G-enablingtechnologies These include the following LTE IEEE 80220 80216e 80216ademMultigigabit Standard WiGig 60 GHz IEEE 80211ad IEEE 1905 Bluetooth v42 andLoWPAN The 5Glab includes network hardware and software computer chips spectrom-eters and cloud computing services The requirements to the 5th generation according to
62 A New Generation of Mobile Communication 227
WirelessCommunication
Automation
loT
Big Data andHPC
OperatingSystems
Audio amp HapticEngineering
Safety Privacyand Security
System-on-a-Chip Integration
SoftwareEngineering
Databases
NetworkedEmbeddedSystems
Human-Machine-Interfaces
Storages
CommunicationTheory
AntennasRF and
PhotonicsEngineering
Fig 613 Requirements to 5th generation according to the 5Glab in Dresden
the visions and initial findings of the 5Glab [212] are given in Fig 613 Nowadays mobilecommunication is occupied with provisioning in general of IP services and transmission ofmultimedia content from one place to another But tomorrow a new generation will be ableto control a wide range of objects in real time with only insignificant human intervention inthe frame of IoT It is necessary to optimise existing systems and mobile wireless networksparticularly in terms of data rate latency interference and reliability according to the staffof the 5Glab
The intentions of 5glabde in Dresden are depicted in Fig 614 Based on the sentencethat ldquo The Internet will disappear in our senses and sensitivitiesrdquo (by E Schmidt) wecan constitute that in opposite to it the future Internet will become 5G Tactile InternetThe breakthrough requirements characterise this transformation into the new 5G TactileInternet with advanced QoS parameters 10 Gbps 1 ms RTT 10000 sensors per cell 108
less outage as well as more security and heterogeneity
Huawei and 5G radio mobile Huawei Technologies was founded in 1987 and actsnowadays as one the largest telecommunications equipment and handset manufacturersin the world By the opinion of Huawei there are the three major design objectives for 5G
1 Implementation of ldquomassive capacityrdquo and ldquomassive connectivityrdquo (similar to theprevious vision)
228 6 Future Mobile Communication From 4G To 5G 5G Enabling Techniques
Fig 614 The intentions of 5glabde towards 5G Tactile Internet
2010 2011 2012
5G Research Prototype Trial
Rel 10 Rel 11
LTE-Advanced
IMT New Spectrum Vision Requirement Technology Eval
LTE-B LTE-C
3GPP
ITU
Rel 12 Rel 13 Rel 14 Rel 15 Rel 16
5G Standard Product Deployment
2013 2014 2015 2016 2017 2018 2019 2020 2021
Time
5G
Fig 615 5G roadmap according to Huawei (Source huaweicom5gwhitepaper)
2 Flexible and efficient use of all available spectra for different network deploymentscenarios (refer to the DIDO concept)
3 An adaptive network solution framework will become a necessity for accommodatingboth LTE and air interface evolution Results from research on clouds and software-defined networks will reshape the entire mobile ecosystem The possible 5G roadmapcan be realised as follows according to Huawei (Fig 615)
As one can see the efforts for 5G development are running in parallel to the deploymentof the new releases for 4GLTE up to LTE-C release 16 [4] The new developmentfor all-spectrum radio access nodes will require the achievements in fundamental radio
62 A New Generation of Mobile Communication 229
F in GHz
Europe
300 150
02 025 05 10 2 3 4 6 810 20 40 60 100
IR UV
60
A B C D E F G H I J K L M
30 15 75 5 3 15 075 05 03 000005λ in cm
Fig 616 5G radio frequency bands EndashL
Fig 617 The Huawei 5G network integrated architecture (300 MHz up to 300 GHz in themid-term) within an IoT
technologies like the air interface RAN radio frequency transceiver and devices Thecontext for the typical radio frequency bands is to be deployed or licensed for thefuture 5G mobile radio networks including the bands EndashL It is depicted in Fig 616It means primarily the broad frequency span 2ndash60 GHz The wave lengths are placedcorrespondingly between 15 and 05 cm
The advanced radio backhaul and new fiber access for the fixed network will be an inte-gral part of next generation commercial network solutions within 5G The interoperabilitywithin 5G network architecture as well the future extension of the since-used 3G cellhierarchy (according to Huawei) is depicted in Fig 617 The Tera-cells are foreseen withthe backhauls to the usually existing 3Gndash4G macro- and microcells The interoperabilitywith the fixed part enables data rates up to 100 Tbps