APAN Cloud WG (2015/3/2)

  • Published on
    14-Jul-2015

  • View
    205

  • Download
    0

Embed Size (px)

Transcript

  • Building High-Performance Inter-Cloud Infrastructure in Japan

    Masaharu Munetomo

    Professor & Vice Director,

    Information Initiative Center,

    Hokkaido University, Sapporo, JAPAN.

    munetomo@iic.hokudai.ac.jp

    HOKKA IDO UN IVERS ITY GU IDEBOOK 2014-2015

    46Campus Maps

    Campus Maps

    1

    2

    3

    4

    5

    6

    7

    8

    9

    10

    11

    P

    P

    P

    P

    P

    To Hakodate Station

    N

    20

    19

    18

    17

    16

    15

    1413

    12

    11

    10

    9

    8

    7

    8

    6

    5

    4

    3

    2

    1

    ... we are cosmopolitan, and accessible...

    Picturesque Hakodate is home to Hokkaido Universitys Faculty of Fisheries Science and is located on the south-west of the island.

    With a population of approximately 280,000 people, the coastal city is at the base of Mount Hakodate, which boasts amazing natural beauty. The

    view from the summit is renowned for having one of the most beautiful views in Japan, particulary at night. Since it opened in 1935, the Hakodate

    Campus has had a proud history of being at the forefront of fisheries research and is one of the largest campuses of its kind in Japan.

    Main Gate Main Building Annex BuildingMarine Bioresources Research Building Marine Frontier Research Building Marine Science Creative Research Building Lecture-room BuildingStudent Laboratories Controlled Environment RoomsTowing Tank RoomAuditorium

    Hakodate Campus Map

    Library Aquatic Biological Specimen House (Nakabe Hall)Fisheries Museum (main building) Fisheries Museum (annex) Gymnasium Student Center Student Activities Building Swimming Pool Athletic Field

    12

    13

    14

    15

    16

    17

    18

    19

    20

    Sapporo Campus

    Hokkaido

    Hakodate Campus

    1

  • Masaharu Munetomo

    Professor & Vice director, Information Initiative Center,Hokkaido university, Sapporo, JAPAN.

    Chief examiner, Cloud computing research group of national supercomputing centers in Japan.

    Chief examiner, SIG Cloud, Academic eXchange for Information Environment and Strategy (AXIES) in Japan.

    Chief examiner, SIG Mathematical Problem-Solving, Information Processing Society of Japan (IPSJ)

    General advisor, Cloud Utilization Promotion Agency (CUPA) & Managed Service Providers associations in Japan (MSPJ)

    Founding member and of steering committee, Open Compute Project in Japan (OCPJ)

    2

  • Information Initiative Center, Hokkaido University

    Founded in 1962 as a national supercomputing center.

    A member of High Performance Computing Infrastructure (HPCI) and Joint Usage/Research Center for Interdisciplinary Large-scale Information Infrastructure (JHPCN) in Japan.

    University R&D center for Supercomputing, Cloud computing, Networking, IT systems for education

    Supercomputer (172TFlops) & Academic Cloud System (43TFlops)

    @(*[( %!*.97W1)'*%1LXN"*2G[@(*89=)+Y=)H4:3RNEIV[57JMO?A/-Q0KUF@(*6T

    3

  • HPCI (High Performance Computing Infrastructure)

    Collaboration of national supercomputing centers in Japan.

    RIKEN AICS (K computer) & Supercomputing Centers (University, Research Institutes) connected via academic high-speed network (SINET4)

    Federations of users & systems management (GSI-SSH, Gfarm supported)

    http://hpci-office.jp/4

  • Hokkaido University Academic Cloud System

    Largest Academic Cloud System in Japan started services from Nov. 2011: 43TFlops (5,000 cores), and more than 2,000 VMs can be deployed.

    Employing CloudStack to provide cloud management portal. High-performance cloud system: each physical node has 40-

    cores, 128GB memory. Network: 10GbE x 2, Shared Storage: 260TB (SAN) + 500TB (NAS) + 2PB (WebDAV, S3, Gfarm)

    Hitach BladeSymphony BS2000 Xeon E7 8870 2.4GHz (10-core) x 4

    128GB memory / 10GbE x 2

    Hitachi NAS Storage AMS2300: 260TB AMS2500: 500TB 5

  • Use case: Big Data processing systems

    We provide Big Data service VM package consisting of Hadoop, Hive, Mahout, and R.

    Automated deployment of VM clusters, customizing scheduling policies in CloudStack to balance I/O overheads for cluster packages (Hadoop / MPI / Torque).

    Storage #3Virtual(Disk

    Storage #4Virtual(Disk

    Storage #2Virtual(Disk

    Zone!POD!

    Shared Storage #1Resource Pool #1

    HyperVisor #2

    HyperVisor #1Virtual(DiskVM(

    Balancing!overheads!of!disk!I/O!with!round8robin!assignment!of!Virtual!disks.!

    Storage #1

    VM(

    VM(

    VM(

    VM(

    Virtual(DiskHadoop Cluster

    Shared Storage #2Resource Pool #2

    HyperVisor #4

    HyperVisor #3

    Virtual(Disk

    VM(

    Shared Storage #3Resouce Pool #3

    HyperVisor #6

    HyperVisor #5

    Virtual((Disk

    VM(

    Shared Storage #4Resouce Pool #4

    HyperVisor #8

    HyperVisor #7

    Virtual(Disk

    VM(

    6

  • Use case: simulation environment to replace in-house computing servers or clusters

    Replacement of in-house clusters of laboratories employing L (10-core) or XL (40-core) project servers.

    Filling in the gap between PCs and super-computers.

    7

  • Use case: development of in-silico screening system for drug design

    Center for Research and Education on Drug Discovery builds a Structure Based Drug Design (SBDD) system for in-silico screening with the academic cloud system

    A virtual private cloud system using XL servers (40-core): modeFRONTIER and AutoDock are installed as docking applications.

    AutoDock[1]

    AutoDock[2]

    AutoDock

    AutoDock

    AutoDock

    AutoDockContinuous execution of analysis

    servers

    8

  • Use-case: Fishing ground prediction system

    Researchers in department of fishery build a fishing ground prediction system on Hokkaido university academic cloud system

    The system provides information on promising sea area for fishing boats to catch squids, employing satellite images and data assimilation results.

    Portal System

    Satellite image processing Data assimilation

    Fishing ground prediction

    INMARSAT

    Satellite Earth station

    Satellite Communications

    Squid Fishing Boats Fishing ground prediction system portal

    9

  • Use-case: Employing PaaS for scalable interactive evolutionary computation

    Building a scalable interactive evolutionary computation framework to evolve solutions according to the preferences of millions of users.

    CloudStack

    VM Ubuntu

    instance

    VM Ubuntu

    Redis

    VM Ubuntu

    Redis

    VM Ubuntu

    Redis

    Database

    VM Ubuntu

    instance

    VM Ubuntu

    instance

    Applycation resource

    iGA iGA iGA

    Load Balancer

    CloudFoundry

    Sever

    Interactive Evolutionary Computation using PaaS

    Users select solutions according to their preferences

    Present cadndates of solutions from the system

    10

  • Japanese academic inter-cloud infrastructure

    Development of the inter-cloud system over Japanese universities to collaborate private clouds from Kitami (Northernmost) to Ryukyu (Southernmost) universities through Japanese academic high-speed network (SINET4).

    Hokkaido University

    Kitami Institute of Technology

    University of Ryukyus (Okinawa)

    National Institute of Informatics (NII)

    @(*[( %!*.97W1)'*%1LXN"*2G[@(*89=)+Y=)H4:3RNEIV[57JMO?A/-Q0KUF@(*6T

    11

  • Related projects

    Remote collaborations of distributed cloud systems (JHPCN)

    Federations technologies development toward academic inter-cloud (Collaborative research project, National Institute of Informatics)

    Large-scale Distributed Design Exploration Framework (JHPCN)

    Development of distributed database infrastructure across Japan

    Inter-cloud resource optimization with multi-objective evolutionary algorithms

    Designing the next-generation Hokkaido university high-performance inter-cloud system

    12

  • Remote collaborations of distributed cloud systems

    Prototyping an inter-cloud manager and authentication infrastructure for federation of academic cloud systems managed by dierent cloud middleware (CloudStack, OpenStack, etc.)

    Designing a VPC (Virtual Private Cloud) management framework in the distributed inter-cloud systems.

    Cloud A IaaS

    Cloud B IaaS

    Cloud C IaaS

    User

    VPC 1

    Internet

    VMVM

    VM

    VPC 2

    220km

    13

  • Large-scale Distributed Design Exploration Framework (LDDEF) To establish a framework to support parameter surveys by

    supercomputing simulations collaborating design engineers sharing information on promising solutions with distributed DBs

    Multi-objective designexploration exploresPareto-fronts stored indistributed DBs

    Optional info. Isstored in objectstorages forvisualization andanalysis Solutions DB (distributed)

    Automated replication for DR and

    load balancing

    Visualization

    Simulation (Supercomputer)

    Optimization & DB management (Cloud system)

    Distributed Database

    Product

    14

  • Grid Unified Framework for Optimization (Grid-UFO) & MHGRID (Asim, Wahib, Munetomo, 2008-2010)

    A unified framework collaborating optimization algorithms libraries and simulation programs to evaluate fitness values registered by dierent developers in GRID computing distributed exec. environment.

    GridUFO(Checks(compa3bility(of(sovler:obj(func(pair(

    (Solvers(Database(

    (Obj(Func(Database(

    User(Develops(&(Registers(an((Op3miza3on(Problem(

    User(Develops(&(Registers(a(Solver(

    Solver(Developer(

    User(Selects(a(Solver(&(an(Objec3ve(Func3on(

    GridUFO(Deploys(the(Job(over(Grid(

    Solver(Obj(Func(

    MHAPI(

    Ninf:IDL(

    Distributed(Implementa3on(over(Grid(

    MHML(

    Obj(Func(Developer(

    Ordinary(User(

    Subm

    its(Op3

    miza3

    on(Jo

    b(

    MHML(

    15

  • LDDEF: System architecture overview

    Fully distributed and scalable architecture consisting of simulators in supercomputers, optimization engines, analyzers object storages and distributed database nodes in the inter-cloud environment.

    DB

    Object)Storage(s)

    DB

    DB

    Simulator

    Optimizer Simulator

    Optimizer

    replication

    {::}

    {::}

    Analyzer:/ Visualizer)

    Controller:&)User:Interface

    Distributed:DBs

    {::}

    {::} (feedback)

    replication

    16

  • Cassandra distributed database nodes deployed across Japan

    We have built a testbed of Cassandra distributed database nodes across Japan from Kitami (Hokkaido) to Okinawa connected via SDN (Vyatta).

    We have tested performance with/without replications and availability and resiliency in cases of node and network faults.

    0"

    1000"

    2000"

    3000"

    4000"

    5000"

    6000"

    1" 11" 21" 31" 41" 51" 61" 71" 81"

    0"

    1000"

    2000"

    3000"

    4000"

    5000"

    1" 11" 21" 31" 41" 51" 61" 71" 81"

    Num

    ber"of"requestsNum

    ber"of"requests

    write8latency"(ms)

    read8latency"(ms)

    with"replicaCons without"replicaConsHokkaido'University'Informa3on'Ini3a3ve'Center

    Kitami'Ins3tute''of'Technology

    University'of'the'Ryukyus'

    70ms

    60ms

    10ms

    17

  • Cloud Resource Deployment Optimization (CReDO) in the Inter-Cloud Environment Optimizing deployment of virtualized systems requested from

    users according to their system specifications using multi-objective evolutionary algorithms such as NSGA-II/III.

    Semi-automated scheduling policy to recommend a variety of system deployment patterns at Pareto-front to users.

    CReDO Solver / Optimizer

    DB

    Request with Spec. info

    Response with Deploy. info

    Public Cloud A Public Cloud B Private Cloud

    System info., Accounting, etc

    18

  • Multi-objective inter-cloud resource optimization using multi-objective evolutionary algorithms.

    We employ multi-objective evolutionary algorithms such as NSGA-II and NSGA-III to solve resource optimization problems in the inter-cloud environment.

    Solving multi-objective optimization considering cost, performance(response time), and greenness (CO2 emission) simultaneously.

    19

  • Toward the next generation of Hokkaido university academic cloud as high-performance inter-cloud

    We are planning to develop a high-performance inter-cloud system as the next generation Hokkaido university academic cloud

    Inter-cloud (service layer): multi-cloud controller & broker with cloud exchange

    Inter-cloud (infrastructure layer): Inter-cloud connector with SDN controller

    Private Cloud with Supercompter & BigData Storage

    Inter-Cloud Portal (multi-cloud controller)

    Public Cloud A

    VPN (SDN)Public Cloud B

    Inter-Cloud Connector

    Community Cloud CPublic/Comunity Clouds

    Cloud Exchange

    HPC

    20

  • High-performance inter-cloud design: an example

    SW Super'computer40GbE'(x'142)

    FC'or'IB

    Tape'Archive

    IaaS/'HaaS'

    IaaS/'HaaS'

    IaaS/'HaaS'

    Cloud'Shared'Storage

    App WebDAV'S3/Gfarm

    HPC'Storage

    Campus'LAN

    FW'Router

    SW

    Management'servers'

    Baremetal,'VMWare) Baremetal,'CloudStack'or'OpenStack) Baremetal)

    100G'x'1 Tape'ArchiveIaaS/'

    HaaS'IaaS/'HaaS'

    Public'clouds'Community'clouds'

    Remote'site'#1 Remote'site'#2

    SW SW SW SW

    IPS'

    40GbE'(x'142)

    SINET5'

    Campus'DC

    SDN'ontroller

    22

  • Roadmap & Future direction

    2016Q2: Upgrade network infrastructure (SINET5: 100Gbps)

    2017Q2-Q3: Replacing inter-cloud infrastructure (including remote sites) & supercomputer at Hokkaido university

    Regional inter-cloud collaborations in Hokkaido

    National inter-cloud collaborations with other universities, NII and other research institutes to establish academic community cloud federations

    International inter-cloud collaborations (Asia-Pacific?)

    Investigations on future trends in inter-cloud applications such as IoT/IoE, extreme-scale parallel and distributed computing including big data processing and machine learning.

    23

  • CloudWeek2015@Hokkaido University

    A collection of symposium, conference, and workshop related cloud computing technologies, sponsored by information initiative center, Hokkaido University.

    Sep.7th - 9th or 10th, 2015, at Hokkaido University, Sapporo, Japan.

    Academic Inter-Cloud Symposium 2015 for Universities, Research institutes

    Open Cloud Conference 2015 for Cloud service providers, vendors, etc.

    ITRC RICC (Regional Inter-Cloud Committee) Workshop

    Call for international speakers!

    Joint usage/research base

    Cloud Week@Hokkaido University (cloud symposium)

    Promotion of open-type joint research

    Introduction of Petabyte-class Data Science Unified Cloud Storage

    Innovative high-performance computing infrastructure (HPCI)

    Administration organization of the Information Initiative CenterMain activities of the Information Initiative Center

    Cloud Week 2013@Hokkaido University as of Jul. 1, 2014

    61422

    11411161532

    55

    ProfessorSpecially appointed professorAssociate professorAssistant professor(Visiting researcher)Research support promotion workerClerical workerTechnical workerSpecific professionalPart-time workerResearch support promotion workerClerical assistantClerical support worker

    Research divisions

    Total

    Clerical divisions (Information Environment Promotion Headquarters, Information Promotion Division)

    Director of the Information Initiative Center

    University of Tokyo Information TechnologyCenter

    Kyoto UniversityAcademic Center for Computing and Media Studies

    Tsukuba University Center for Computational Science

    Tokyo Institute of Technology Global Scientific Information and Computing Center

    Japan Agency for Marine-Earth Science and Technology Center for Earth Information Science and Technology

    Kyushu University Research Institute for Information Technology

    Osaka University Cybermedia Center

    Nagoya University Information Technology Center

    Hokkaido UniversityInformation InitiativeCenter

    Tohoku UniversityCyber Science Center

    RIKEN Advanced Institutefor Computational Science K

    Institute of Statistical Mathematics Center for Engineering and Technical Support

    Faculty members

    Organization chart Research divisionsAs a nationwide joint use facility, the Center aims to promote the advancement of education and research and the implementation and support of education taking advantage of information media, through R&D to facilitate informatization and the development and operation of an inform...

Recommended

View more >