Upload
thierry-coupaye
View
1.248
Download
0
Embed Size (px)
Citation preview
1 © Orange
Towards the extinction of mega data centres? To which extent should the Cloud be distributed at the network edge?
Keynote @ 4th IEEE International Conference on Cloud Networking, Niagara Falls, Canada
October 5-7 2015
Thierry Coupaye (PhD)
Head of cloud platforms research, Orange Labs, France
2 © Orange
§ Cloud evolutions, variations and mutations
§ Centralized or distributed: pick one!
§ Centralized or distributed: take both!
§ End of story: natural selection?
Outline
3 © Orange
The speaker
§ Head of cloud platforms research at Orange Labs
§ Orange Expert on Future Network - Sponsor of the “Programmable and cloud networking” domain
§ Background in distributed systems architecture - Active database systems at University of Grenoble, France - Semi-structured data management at European Bioinformatics Institute,
Cambridge, UK - Large scale software deployment at Dassault Systems, France - Component-based software architectures at Orange, France
-> autonomic and cloud computing
5 © Orange
Why is Cloud Computing strategic for Orange
’’Cloud Computing is in our DNA’’. Orange Business Services, Dec.
2009
“Cloud Computing is here to drive the IT ecosystem so traditional ICT
providers must transform or die.” Yankee Group, 2008
“Cloud computing holds enormous potential for telecom service
providers if they get aggressive about driving technological
innovation there.” Telephony Online, April 2009.
1. evolution of hosting offers for enterprise market XaaS for multinational companies, large national accounts,
and medium and small enterprises
2. evolution of mass market services platforms and services - communications, audiovisual (TV, DVR, music), healthcare,
IoT/M2M, transport, gaming… - emergence of personal cloud
3. evolution of information systems – network and service provisioning and management,
customers management, business intelligence, billing… – human resources, inventory (network, suppliers),
finances…
4. embodiment of Future Internet Architecture “SDN”, “NFV”, “Network Softwarization”
§ Very few networks/platforms/services inside Orange will not be impacted by cloud computing
§ The future of Orange depends partly on its capability to master cloud computing
8 © Orange
Back to the definition of Cloud Computing (adapted from NIST)
6 characteristics
1. On Demand
2. Self-Service
3. Network Access
4. Resource Pooling (multi-tenancy)
5. Rapid Elasticity
6. Measured Service, Scalable pricing
3 delivery models
(markets) 1. Cloud Software
as a Service (SaaS)
2. Cloud Platform as a Service (PaaS)
3. Cloud Infrastructure as a Service (IaaS)
4 deployment models
1. Private cloud
2. Public cloud
3. Hybrid cloud
4. Community Cloud
Source: Forrester
Location transparency is the point!
9 © Orange
“The future of cloud computing - an army of monkeys?” Sam Johnston, Sept. 2008
“I don't care if my cloud computing architecture is powered by a grid, a mainframe, my neighbour's desktop or an army of monkeys, so long as it's fast, cheap and secure.”
Source: https://groups.google.com/forum/#!topic/cloud-computing/xs2ctSEvMbw
10 © Orange
Trends, variations, mutations towards cloud distribution Centralized public clouds are in fact generally distributed over multiple (mega) data centres for availability reasons
Verizon (©)
Orange (©) Microsoft (©)
Amazon (©)
1
14 © Orange
Orange Data Centres
Sydney core MPLS POP
other core backbone routes
data centres
core DC interconnections
Atlanta, GA
Normandie
Lodz
Rio de Janeiro
major service centres
Mauritius
Cairo Delhi
London
San Jose, CA
Sterling, VA
Francfurt
Singapore
Hong Kong
Paris
15 © Orange
Hybrid and community clouds are by nature distributed over multiple data centres/clouds
2
© Avaeglo
© Orange
Trends, variations, mutations towards cloud distribution
16 © Orange
Networks are getting « softwarized » and are converging with a distributed vision of cloud computing.
3 examples: § Virtual CDN (vCDN)
§ Cloud RAN (C-RAN)
§ Mobile Edge Computing (MEC)
3
© Dataquest
Trends, variations, mutations towards cloud distribution
17 © Orange
Virtual Content Delivery Networks (vCDN)
§ vCDN is a step towards CDN “cloudification”
§ vCDN is a step towards cloud distribution at the edge of the network CDN providers would like to deploy their functions inside mobile networks
© Akamai
© Amazon
§ CDN - Caching servers placed closed to users - Used for video streaming and also web
accelaration, device management, application delivery, virtual desktops…
- Less effective in varying conditions (e.g. flash crowds)
§ vCDN – virtualization-based CDN - Dynamicity of ressource (caches)
location - Isolation in multi-providers scenarios
18 © Orange
Cloud Radio Access Networks (C-RAN)
§ C-RAN is a step towards RAN “cloudification”
§ C-RAN is a step towards cloud distribution at the edge of the network in the access network because RRH and BBU distance cannot exceed ~20-40kms
§ Traditional/All-in-One/Macro-Base architecture - Co-location of radio and baseband processing
§ Distributed Base Station - Separation of radio (RRH) and baseband processing (BBU) - More convenient place for BBU (maintenance…) - Still static assignment of BBU to RRH
§ Cloud-RAN
- BBU virtualization and pooling Dynamicity/flexibility in RRH-BBU assignment, better performance, energy and cost savings, easier maintenance and evolution
Source: “Cloud RAN for Mobile Networks – A Technology Overview”. A. Checko and al. IEEE Communication Surveys & Tutorials, 7(1), 2015.
19 © Orange
Mobile Edge Computing (MEC)
§ MEC is a step towards cloud distribution at the edge of the network
§ MEC targets geo-distributed applications
§ ETSI ISG targetting: - a standard service environment open
to 3d party service developpers and content providers
- on top of a standard hosting infrastructure
- located at the RAN Edge
§ Benefits: low latency, high bandwidth, access to radio network data
§ Use cases: video analytics, location-based services, IoT, augmented reality, caching
© ETSI
21 © Orange
Economical issues
Criteria Centralized Cloud/Mega DC Distributed Cloud/Micro DC
Construction cost (terrain, building)
Complex to choose a site, to get authorisations.
Smaller and more discrete buildings. Existing buildings can be reused.
Extension cost Very low up to the DC physical limit. Very high beyond: need to build new mega DC.
Small and smooth. Very low until the DC physical limit. Easy beyond to rent or build a new small building or a « DC container »
Construction security cost
Strong, active, humanized security. A mega-DC is an industrial site…
Can be passive and light
😊 😟
😊 😟
😊 😕
© Orange
© Sun Microsystems
22 © Orange
Environmental issues
Criteria Centralized Cloud/Mega DC Distributed Cloud/Micro DC Energy cost Good price because of high
volumes but heat reselling may be difficult for isolated DC
Less or no need for cooling. Higher capacity to use locally produced clean energy (solar, wind…). Easyness to resell heat to surrounding buildings
Energetic connectivity
DC needs to be close to massive electricity production. Redondance of energy producers may be difficult (e.g. in France)
No special approvisionnement. Easyness to get multiple producers
Network connectivity
Large DC can be located close to big peering points
Most micro DC are not always close to major peering points
DC workers transportation cost
Workers generally not in close proximity
May need transportation but workers live not far away
Risks (industrial, natural, economic…)
Large size implies higher risks and associated cost
Lower risks. Lower cost.
😊😟
😊😕
😏 😏
😊 😟
😟 😊
© Qarnot Computing
23 © Orange
Technical issues
Criteria Centralized Cloud/Mega DC Distributed Cloud/Micro DC
Manageability (supersvision)
Centralized management is easier (homogeneity)
More complex, need more automation but hardware management easier.
Security Easy security management. Smaller attack surface (risk fragmentation). Less damage in case of attacks.
Availability and reliability
SPOF risk. Difficult DC redundancy. Higher dependability on core network and Internet traffic
No SPOF. Easier recovery. Inter-DC fault tolerance. Local traffic.
Performance Very efficient « inside the DC » but not necessarily from user’s point of view.
Lower latencies. Intrinsic support of user mobility (follow-me cloud)
Core network traffic
Heavy Lighter
😕 😏
😊😏 😟😊
😊😟
😟
© University of Technology Sydney
😊
24 © Orange
Socio-political issues
Criteria Centralized Cloud/Mega DC Distributed Cloud/DC
Proximity and data protection
Concerns with data location in remote/foreign mega DC
More trust in local data centre (e.g. “proxicenter” in Rennes, France)
Sovereignty and land-use planning
Huge economic attraction but limited to happy few and mega DC generally out of city centres
Local authorities (e.g. cities) expectations i) from their own needs (smart cities) and ii) to attract business
Legal issues Concerns with data location (e.g. health, fiscal), lawful interception
Easier adaption to local regulations. QoE improvement through data location (cf. Net Neutrality)
“Libertarianism”
Embodiment of “digital imperialism”
Contributes to fulfilment of expectations for decentralized and open (source) infrastructures
Proxycenter, Rennes, France © TDF
😊
😊
😊
😊
😟
😕
😟
😟
26 © Orange
Fog Computing
§ A paradigm from Cisco “Fog Computing is a highly virtualized
platform that provides compute, storage and networking services between end devices and
traditional Cloud Computing data centres, typically, but not exclusively located at the
edge of the network” *
§ Cisco Fog Computing = Cloud + IoT - Data collection from sensors/things - Local data processing and actuators
control - Filtering, aggregation and upload to
remote DC for batch analysis
§ Use cases: connected vehicles, smart grid, wireless sensors and actuators… web acceleration
© Cisco
* Source: “Fog Computing and Its Role in the Internet of Things”. Flavio Bononi and al, Cisco. ACM SIGCOMM International Conference on Mobile Cloud Computing, August 2012.
27 © Orange
Some other geo-distributed clouds § NTT Edge Computing - Small servers in vicinity of users
and devices - Uses cases: smart city/building,
M2M, medical, gaming, speech/image recognition
- Edge accelerated web platform research prototype
§ AT&T Cloud 2.0 - Balance between local storage/
computation and remote offloading
- Favor local storage/computation by a device or a federation of devices (~Device-to-Device)
§ SAVI - Canadian initiative: U. Waterloo,
Toronto, McGill… + IBM, Cisco, Juniper…
- Testbed • for accelerated distributed
(possibly short-lived) applications deployment
• over virtualized small-cell wireless access network
• Connected through optical backhaul to multi-tier cloud including both mega DC and smart converged edges
28 © Orange
Orange geo-distributed cloud
§ NGPoP - Virtualized converged access network - Virtualized computing and storage - Open to 3d parties
§ Discovery (http://beyondtheclouds.github.io)
- A open initiative lead by Inria with Orange and Renater
- Targets a Locality-based Utility Computing platform (“LUC-OS”)
- Hypothesis • Autonomic and decentralized
management • OpenStack substrate
§ An ubiquitous cloud platform that leverages a continuum of DC from mega DC to nano/pico DC (user devices/things) through mini/micro DC (network PoP)
§ Orange as a (geo) distributed open cloud platform operator
29 © Orange
Geo-distributed cloud for new applications Source: NGMN 5G White Paper
§ “intrinsically local” § Crowd/social § Connected cars § Smart Home § Smart City § IoT in general
§ Interactive § Gaming § E-health § Augmented reality § Virtual reality
31 © Orange
1. Geo-distributed applications will continue to grow
2. The construction of new massive mega DC might slow down
3. Smaller DC in closer users’ vicinity will complement mega DC
4. User devices are potential cloud platforms too (“nano DC”, “CloudLets”)
5. Different cloud deployment scenarios will probably co-exist
- which raises many interesting technical/research challenges…
6. This new landscape could change the actors play
webcos <-> cloud providers <-> CDN providers <-> telcos
Final word: natural selection?
32 © Orange
Thank you!
Acknowledgments This talk contains material from Ivan Meriau, Arnaud Diquélou, Daniel Stern from Orange Labs, Sylvain Quief and al. from Orange Cloud for Business
Link: Please comment on Orange research blog here: http://research.orange.com/en/fog-computing-and-geo-distributed-cloud/