Cost-Benefit Analysis of Cloud Computing versus Desktop Grids

Transcription

1 Cost-Benefit Analysis of Cloud Computing versus Desktop Grids Derrick Kondo, Bahman Javadi, Paul Malécot, Franck Cappello INRIA, France David P. Anderson UC Berkeley, USA

2 Cloud Background Vision Hide complexity of hardware and software management from a user by offering computing as a service Benefits Pay as you go Scale up or down dynamically No hardware management, less software management

3 Platform Performance vs. Costs high Cost low low Performance high

4 Platform Performance vs. Costs high Supercomputer Grid Cluster Cost low low Performance high

5 Platform Performance vs. Costs high Supercomputer Grid Cluster Cost low low Performance Desktop Grids (DG) high

6 Platform Performance vs. Costs high Supercomputer Grid Cluster Cost low low Cloud Performance What is the relationship? Desktop Grids (DG) high

7 Platform Performance vs. Costs high Supercomputer Grid Cluster Cost low low Cloud Performance What is the relationship? Desktop Grids (DG) DG + Cloud? high

8 Outline Performance tradeoffs Monetary tradeoffs Client hosting Server hosting

9 Apples to Apples Loosely-coupled, high-throughput, compute-intensive applications Tightly-coupled, data-intensive real-time applications low complexity high complexity

10 Apples to Apples Loosely-coupled, high-throughput, compute-intensive applications Tightly-coupled, data-intensive real-time applications low complexity high complexity DG s

11 Apples to Apples Loosely-coupled, high-throughput, compute-intensive applications Tightly-coupled, data-intensive real-time applications low complexity high complexity DG s Clouds

12 Apples to Apples Loosely-coupled, high-throughput, compute-intensive applications Tightly-coupled, data-intensive real-time applications low complexity high complexity DG s Clouds Comparison assuming embarrassingly parallel, compute-intensive applications

13 Method Use real performance measurements Exported BOINC (middleware for desktop grids) project data Use real costs Large/small BOINC projects / XtremLab) Amazon Elastic Computing Cloud (EC2)

14 Platform Construction Stages of Project &

15 Stages of Project & Platform Construction Deployment

16 Stages of Project & Platform Construction Deployment Execution

17 Stages of Project & Platform Construction Deployment Execution Completion

18 Platform Construction Deployment Execution Completion How long before I get X TeraFLOPS?

19 Platform Construction Deployment Execution Completion How long before I get X TeraFLOPS? Number of cloud nodes TeraFLOPS Months for registration

20 Platform Construction Deployment Execution Completion How long before I get X TeraFLOPS? Number of cloud nodes Can get over 20 TeraFLOPS within 6 months TeraFLOPS Months for registration

21 Platform Construction Deployment Execution Completion How long before I get X TeraFLOPS? Strategy: Add to BOINC project list Press releases Forum Announcements Google Ad Sense Respond to users (leverage volunteers) Number of cloud nodes Can get over 20 TeraFLOPS within 6 months Months for registration TeraFLOPS

22 Platform Construction Deployment Execution Completion How long to deploy my batch of tasks needing faster response time?

23 Platform Construction Deployment Execution Completion How long to deploy my batch of tasks needing faster response time? 7,82/9*,34.4:*,.;*:3)4,6<!" '!" &!"!!" ".!"!!.!"".4=6>6!""".4=6>6!"""".4=6>6!" #!" $!" % ()*+,-./0.1/2)34,,-.3/5,6

24 Platform Construction Deployment Execution Completion How long to deploy my batch of tasks needing faster response time? 7,82/9*,34.4:*,.;*:3)4,6<!" '!" &!"!!" "!"!!.!"".4=6>6!""".4=6>6!"""".4=6>6 For 1000 tasks, ~10 minutes with 10 5 hosts!" #!" $!" % ()*+,-./0.1/2)34,,-.3/5,6.

25 Platform Construction Deployment Execution Completion How long to deploy my batch of tasks needing faster response time? Strategy: Specify lower latency bounds [Heien et al.] 7,82/9*,34.4:*,.;*:3)4,6<!" '!" &!"!!" "!"!!.!"".4=6>6!""".4=6>6!"""".4=6>6 For 1000 tasks, ~10 minutes with 10 5 hosts!" #!" $!" % ()*+,-./0.1/2)34,,-.3/5,6.

26 Platform Construction Deployment Execution Completion

27 Platform Construction Deployment Execution Completion How many volunteer nodes are equivalent to 1 cloud node?

28 Platform Construction Deployment Execution Completion How many volunteer nodes are equivalent to 1 cloud node?

29 Platform Construction Deployment Execution Completion How many volunteer nodes are equivalent to 1 cloud node? 2.8 active volunteer hosts per 1 cloud node. (Total performance still orders of magnitude better)

30 Platform Construction Deployment Execution Completion How many volunteer nodes are equivalent to 1 cloud node? 2.8 active volunteer hosts per 1 cloud node. (Total performance still orders of magnitude better) Strategy: Use statistical prediction of availability

31 Platform Construction Deployment Execution Completion

32 Platform Construction Deployment Execution Completion How long should I wait for task completion?

33 Platform Construction Deployment Execution Completion How long should I wait for task completion?

34 Platform Construction Deployment Execution Completion How long should I wait for task completion? Median project latency bound: 9 days for 3.7 hour work unit (on 3GHz host). Ratio of lat. bound / exec time > 5. Good success rates: 96.1% of WCG tasks met out of 227,000 tasks

35 Platform Construction Deployment Execution Completion How long should I wait for task completion? Strategy: See BOINC Catalog for typical deadlines and compute/comm/mem ratios. Median project latency bound: 9 days for 3.7 hour work unit (on 3GHz host). Ratio of lat. bound / exec time > 5. Good success rates: 96.1% of WCG tasks met out of 227,000 tasks

36 Monetary Tradeoffs Client hosting on cloud Not worth it and never will Server hosting on the cloud Possible solution

37 Monthly Project Costs

38 EC2 Pricing

39 What about Client Hosting on the Cloud?

40 Cost of Clouds versus DG How long until DG s more cost effective than Clouds?

41 Cost of Clouds versus DG How long until DG s more cost effective than Clouds? 7,8./9,)/:;<( #!!## #!!#$ #!!#% #!!#' F#'!'/G*HI8.,,)/8*3,5 2'J&/-H*I3/8*3,56 '"&$/G*HI8.,,)/8*3,5 2#&#$/-H*I3/8*3,56 A7$ / #!!#" #!!#& /! " #! #" $! $" %! ()*+,-./01,/230456

42 Cost of Clouds versus DG How long until DG s more cost effective than Clouds? 7,8./9,)/:;<( #!!## #!!#$ #!!#% #!!#' F#'!'/G*HI8.,,)/8*3,5 2'J&/-H*I3/8*3,56 '"&$/G*HI8.,,)/8*3,5 2#&#$/-H*I3/8*3,56 A7$ / #!!#" After ~13 days #!!#& /! " #! #" $! $" %! ()*+,-./01,/230456

43 Cost of Clouds versus DG (2) What are total costs over time?

44 Cost of Clouds versus DG (2) 8 x 104 What are total costs over time? Total cost EC2 100n EC2 1000n EC n XtremLab SETI@home Platform age (days)

45 Cost of Clouds versus DG (2) 8 x 104 What are total costs over time? Total cost After 3 days, DG is cheaper EC2 100n EC2 1000n EC n XtremLab SETI@home Platform age (days)

46 Cost of Clouds versus DG (3) How many months of DG can X months of Cloud buy me?

47 Cost of Clouds versus DG (3) How many months of DG can X months of Cloud buy me? Months with VC Xtremlab / 1000n EC2 SETI@home / 1000n EC2 Xtremlab / 10000n EC2 SETI@home / 10000n EC Months with EC2

48 Cost of Clouds versus DG (3) How many months of DG can X months of Cloud buy me? Months with VC Xtremlab / 1000n EC2 SETI@home / 1000n EC2 Xtremlab / 10000n EC2 SETI@home / 10000n EC2 12 months of Cloud can pay for > 125 DG years. DG way more sustainable Months with EC2

49 Equivalent Clouds given BOINC Costs Given BOINC costs, what size Cloud can I buy?

50 Equivalent Clouds given BOINC Costs Given BOINC costs, what size Cloud can I buy?

51 Equivalent Clouds given BOINC Costs 1.75 TeraFLOPS, 7 TeraBytes Given BOINC costs, what size Cloud can I buy?.75 TeraFLOPS, 3 TeraBytes

52 Equivalent Clouds given BOINC Costs Given BOINC costs, what size Cloud can I buy? 1.75 TeraFLOPS, 7 TeraBytes 2 orders of magnitude lower than BOINC.75 TeraFLOPS, 3 TeraBytes

53 What about Server Hosting on the Cloud?

54 Number of Hosts over Time

55 Number of Hosts over Time Load variation exits w/ publicity, projects run out of work, etc. Clouds take care of server management. Potential to exploit clouds.

56 Server Costs on a Cloud How much to host BOINC server on cloud?

57 Server Costs on a Cloud How much to host BOINC server on cloud?

58 Server Costs on a Cloud How much to host BOINC server on cloud? (versus 43K +12K/month) (versus {4,1}K + {5,1}K/month)

59 Server Costs on a Cloud How much to host BOINC server on cloud? times cheaper to host on Cloud, but bandwidth is expensive. (versus 43K +12K/month) (versus {4,1}K + {5,1}K/month)

60 Cloud Resources with Given Budget How big of a server can I support with given budget?

61 Cloud Resources with Given Budget How big of a server can I support with given budget?

62 Cloud Resources with Given Budget How big of a server can I support with given budget? 4 Mbps, 2 TB 12 Mbps, 2 TB

63 Cloud Resources with Given Budget How big of a server can I support with given budget? Many project servers are sustainable on the cloud 4 Mbps, 2 TB 12 Mbps, 2 TB

64 Summary

65 Performance tradeoffs Summary 20 DG TeraFLOPS within 6 months

66 Summary Performance tradeoffs 20 DG TeraFLOPS within 6 months Monetary tradeoffs Client hosting After 13 days, DG more cost effective Server hosting DG server on cloud is cost-effective Best for small-medium sized projects Savings of at least 40%

67 Summary Performance tradeoffs 20 DG TeraFLOPS within 6 months Monetary tradeoffs Client hosting After 13 days, DG more cost effective Server hosting DG server on cloud is cost-effective Best for small-medium sized projects Savings of at least 40% BOINC server image for EC2 / Xen available

68 Future Work Statistical and predictive methods for ensuring host availability Lightweight mechanisms for virtual machine migration

69 wu flop calc bw in cloud how is cloud equiv calculated