クラウドを支えるハードウェア・ソフトウェア基盤技術

  • 3,497 views
Uploaded on

 

More in: Technology
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
No Downloads

Views

Total Views
3,497
On Slideshare
0
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
100
Comments
0
Likes
1

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. 2010 12 8 Photo by digital cat
  • 2. ••• OS• • • TCP/IP •• •
  • 3. • ••• • Amazon Google Microsoft • HPC•
  • 4. Photo by ajlvi
  • 5. 2006 8 9 Search Engine StrategiesConference “ ” ... Wikipedia
  • 6. GoogleSOA 2004 2QWeb2.0 2005 4Q 2006SaaS 2006 4Q 2008 1Q 4Q NTT 2009 7 29
  • 7. AmazonAmazon“Amazons cloud computing will surpass its retailingbusiness,” Larry Dignanhttp://blogs.zdnet.com/BTL/?p=84712007 http://www.technologyreview.com/computing/22606/page3/
  • 8. SaaS UC Berkeley ud SP lo A C Web Web SaaS SaaSWeb ASP SaaS IaaS, PaaS SaaS (Software as a Service)
  • 9. SaaS UC Berkeley ud SP lo A C Web Web SaaS SaaSWeb SaaS IaaS, PaaS s y s dr le un b- Fo Fa NVIDIA Intel, Samsung TSMC
  • 10. Gmail••••• etc... PC
  • 11. ••• * • 2 •• *) Jack Dongarra, “An Overview of High Performance Computing and Challenges for the Future,” HPDC 2009.
  • 12. • •• •
  • 13. • Salesforce.com PaaS• 1• SIer •
  • 14. mixi•• Amazon WS • EC2/S3/Load balancing • CloudFront• Flash/Flex
  • 15. •1 180 /• 7000• Amazon EC2 50 Hadoop 30 5•
  • 16. Animoto.com• CPU• Facebook SNS •• Amazon EC2
  • 17. • •• •• •
  • 18. NIST NIST Definition of Cloud Computing v15 http://csrc.nist.gov/groups/SNS/cloud-computing/
  • 19. On-demand self-serviceBroad network access Resource pooling Rapid elasticity Measured service NIST Definition of Cloud Computing v15 http://csrc.nist.gov/groups/SNS/cloud-computing/
  • 20. ! ! ! ! !
  • 21. ••1 10
  • 22. • • • MTBF* 3 1000 1 1 10 1 100 • e.g., Google RAID ECC• • * Mean Time Between Failures
  • 23. Photo by erix!
  • 24. • •• • • 2 PUE PUE = DC IT * Power Usage Effectiveness
  • 25. • IT•• Sun Black Box 2006 Google 2005 1 1160 / 2003
  • 26. Google GooglePublickey 2009 7 17http://www.publickey1.jp/blog/09/post_36.html
  • 27. Microsoft 4 DC Microsoft112 22.4 IT-PAC: Free cooling 2000 / PUE 1.06
  • 28. DC (2006 Microsoft ) 1000 5 $95/Mbps/ $13/Mbps/ 7.1 $2.20/GB/ $0.40/GB/ 5.7 140 / 1000 / 7.1 M. Armbrust, A. Fox, R. Griffith, A. D. Joseph, R. Katz,A. Konwinski, G. Lee, D. Patterson, A. Rabkin, I. Stoica, and M. Zaharia, “Above the Clouds: A Berkeley View of Cloud Computing,” 2009
  • 29. Google• Google 900 675 •3 45 450 225• Microsoft 0 2006Q1 2007Q1 2008Q1 2009Q1 2010Q1 • 2008 21 8000 30• Amazon • 2009 EC2 4• 2009 52 “Who Has the Most Web Servers,” Data Center Knowledge, May 14, 2009, http://www.datacenterknowledge.com/archives/2009/05/14/whos-got-the-most-web-servers/
  • 30. Amazon EC2Google App Engine Amazon Azure Photo by naixn
  • 31. SaaS (Software as a Service): Service Salesforce.com Google AppsContainer PaaS (Platform as a Service): Google App Engine OS Windows Azure VMM aaS (Infrastructure as a Service): HW Amazon EC2 VMM: Virtual Machine Monitor
  • 32. • SaaS (Software as a Service) • • Google Apps Salesforce CRM• PaaS (Platform as a Service) • • Google App Engine Force.com• IaaS (Infrastructure as a Service) • • Amazon EC2 Blue Cloud
  • 33. Amazon Web Services• 2006 S3 • Amazon * • Web• 2006 8 EC2 • 2008 8• *)
  • 34. Amazon• • • Elastic Compute Cloud (EC2) • CloudWatch • Elastic MapReduce • • Auto Scaling • Virtual Private Cloud (VPC)• • Elastic Load Balancing • CloudFrond •• • Flexible Payments Service (FPS) • SimpleDB • DevPay • Relational Database Service (RDS) •• • Simple Storage Service (S3) • Simple Queue Service (SQS) • Elastic Block Storage (EBS) • Simple Notification Service (SNS) • AWS Import/Export
  • 35. Amazon Elastic MapReduce SNS SQS RDS SimpleDB KVSElastic Load Balancing EBS EC2 S3 CloudFront CDN Edge Location
  • 36. Amazon EC2• OS S3• EC2 launch• (Web service API) • upload S3•1 2007 Opteron 1.0 - 1.2 GHz
  • 37. Regions Availability Zone• Region • 4• Availability Zone Region • • AWS region (US WEST) AWS region (APAC) Zone 1 Zone 2 Zone 1 Zone 2
  • 38. EC2• Small 1 $0.085 x 24 x 365 = $744.6• 1 Cluster Compute 10• $1.6 x 24 x 10 = $384 Instance type ECU Memory (GB) Storage (GB) Arch. Price ($/h) Small 1 1.7 160 32bit 0.085 Large 4 7.5 850 64bit 0.34 Ex-large 8 15 1690 64bit 0.68 High-mem quad 26 68.4 1690 64bit 2.4 High-CPU ex-large 20 7 1690 64bit 0.68 Cluster Compute* 33.5 23 1690 64bit 1.6 *) 10
  • 39. EC2• • 1 0.085 ! 0.095• Windows • 3 4 0.085 ! 0.12• • 1/3 • small 1 • 0.085 x 24 x 365 = 744.6 • 0.03 x 24 x 365 + 227.5 = 489.8
  • 40. EC2• • • • 1/2 1/3 • 0.085 -> 0.029 8/20 • • • •
  • 41. AWS• • PaaS: RightScale Heroku EC2 PaaS • IaaS: Eucalyptus EC2/S3• IaaS PaaS •• •
  • 42. SaaS (Software as a Service): Service Salesforce.com Google AppsContainer PaaS (Platform as a Service): Google App Engine OS Windows Azure VMM aaS (Infrastructure as a Service): HW Amazon EC2 VMM: Virtual Machine Monitor
  • 43. Google Google Web Gmail Maps ... Sawzall App Engine MapReduce BigTable Google File System
  • 44. Google App Engine• Web • Web 3 • Python Java• • • • DB •
  • 45. Google App Engine Google App Engine DB DB Bigtable
  • 46. GAE• • • API • 30 •• SDK
  • 47. • • • CPU 6.5 hour/day $0.10/hour 1GB/day $0.10/GB 1GB/day $0.12/GB 1GB/day $0.15/GB x month 2000/day $0.0001/mail
  • 48. Windows Azure• Windows ASP .NET • • Visual Studio• Microsoft • • Windows Azure platform appliance
  • 49. • IaaS • •• PaaS SaaS http://jp.fujitsu.com/solutions/cloud/ http://cloud.nifty.com/ Softbank http://tm.softbank.jp/business/white_cloud/ IIJ GIO http://www.iij.ad.jp/GIO/
  • 50. HPC HPC• • SaaS !IaaS User• App. • IaaS VM VM VM ! H/W H/W H/W
  • 51. TOP500: LINPACK 2010 11 100 InfiniBand: 80% 90 80 70 10 Gigabit Ethernet: 74% 60% 50 40 Gigabit Ethernet: 54% 30 #231 Amazon EC2 InfiniBand 20 cluster compute instances Gigabit Ethernet 10 10 Gigabit Ethernet 0 0 50 100 150 200 250 300 350 400 450 500 Rank ※ : Rmax : Rpeak
  • 52. AIST Green Cloud Amazon EC2 (AGC) Cluster Instance Intel E5540/2.53GHz Intel X5570/2.93GHz CPU 2 sockets 8 cores 2 sockets 8 coresMemory 48 GB 23 GBNetwork InfiniBnad, 10GbE 10GbE SAS 300GB x2 EBS 20GB HDD (RAID1) Ephemeral 850GB x2 CentOS 5.5 CentOS 5.4 OS BMM, PVM, HVM HVM BMM: Bare Metal Machine PVM: Para Virtualized Machine HVM: Full Virtualized Machine
  • 53. MPI PingPong PingPong 2500 Send 0 1 2000 Recv IB-BMM 2Throughput (MB/s) 1500 15 BMM 1000 500 EC2 PVM HVM 0 16 1k 16k 1M 16M 1G Message size (byte)
  • 54. • Bloss: Block Sakurai-Sugiura • 100 • MPI OpenMP rank 0 rank 0 N 760MB Bcast Reduce 10GB 1GB 1GB Bcast Gather 350MB
  • 55. Bloss: Bloss (2 ranks per node, 4 threads per rank) 120 100Parallel Efficiency (%) 80 IB-BMM BMM 60 EC2 5% PVM HVM 40 BMM 21.06 PVM 22.33 20 HVM 22.66 EC2 20.00 0 1 2 4 8 16 Number of nodes
  • 56. • HPC •• • CPU • I/O • Passthrough•
  • 57. Photo by Nicholas_T
  • 58. ••••••• “Assessing the Security Risks of Cloud Computing,” Gartner http://www.gartner.com/DisplayDocument?id=685308
  • 59. DDoSAPI VPN WAN IO VM NoSQL VMPay-for-use M. Armbrust, A. Fox, R. Griffith, A. D. Joseph, R. Katz,A. Konwinski, G. Lee, D. Patterson, A. Rabkin, I. Stoica, and M. Zaharia, “Above the Clouds: A Berkeley View of Cloud Computing,” 2009
  • 60. •• • •• • CPU
  • 61. e-mail: takano-ryousei@aist.go.jptwitter ID: @oraccha
  • 62. • M.Armbrust, et al., “Above the Clouds: A Berkeley View of Cloud Computing,” 2009• L.A.Barroso and U.Holzle, “The Data Center as a Computer” Google• SACSIS 2010• JJUG 2009 10 8• NTT 2009 7 29