Processor Selection ForOptimum MiddlewarePrice / PerformanceDavid A. Kradakra137 at gmail.com1
Which processoris most cost effectiveto run expensivemiddleware for yourapplication?2
One Page SummaryExpensive middleware:• is priced per core differently for different processors;• its costs overshadow serv...
Processor Selection For Optimum Middleware Price / PerformanceOne Page Summarydakra137 at gmail.comExpensive middleware:• ...
AcknowledgementsI wish to thank:• The Transaction Processing Council andSAP for making it so easy to acquirebenchmark resu...
Key QuestionHow do you take into account“Core Weighting Factors,”“Processor Value Units,” etc.when looking to pick aproces...
ExamplesOracle CoreWeightingFactorRatingsAdvantageExampleProcessor0.25 4xOracle SPARCT30.5 2xIntel XEON75xx0.75 1.33x HP P...
AnswerUse“Performance per Weighted Core”rather than“Performance per Core.”8
Performance perWeighted CoreP/WC = P / (Cores x Weight)[where P often is the maximumTPS rate that meets the responsetime r...
Question #2So what is the most costeffective processor?10
Answer #2aThat really does depend on:• the middleware,• the application, and• its scale.11
Answer #2bThis case studyconsiders:• Middleware:Oracle DBMS• Applications:•TPC-C•TPC-HTPC-H scales:•1,000•3,000•10,000 12
Key PointsMiddleware $ >> Everything else thatvaries e.g. ServerHW & OSThis case study avoided:• Exact pricing, for the re...
TPC-C14
Top 20 TPC-C Results(per ENTERPRISE weighted core)Summary XEON POWER SPARC ItaniumTop 5 TPC-C TPMper weighted core4 1Top 1...
Top 20 TPC-C Results(relative cost per EE weighted core)As of Spring 2011Intel Xeon Processor X5570 2.93GHzIntel Quad-Core...
Key PointDon’t be mislead by the quantities ofbenchmarks!The quantity of entries represents thewillingness of vendors to p...
Top 20 TPC-C Results(per weighted core) (as of Spring 2011)ProcessorCores /ProcessorCache PerProcessorClockRateCachePerCor...
Architecture AnalysisFor TPC-C the Hyper-threaded XEONs come out on topbecause of performance, advantageous core weighting...
Source of Top Xeon’s RatingCache? No, the top three outperformed theXeon’s with larger cache per core and largertotal cach...
21IBM POWER6 - 4.7 GHzIBM POWER5+ - 2.2 GHzSPARC T3 1.65GHzIBM POWER5 - 1.9 GHzIntel Itanium2 Dual-Core - 1.6 GHzIntel Ita...
TPC-H22
TPC-H Results Summary(per weighted core)(as of Spring 2011)Scale10,000Scale 3,000 Scale 1,0001.Itanium(9x40)SPARC (64 VII)...
TPC-H Results Summary(relative cost per weighted core)Intel Itanium Dual-Core 9140 - 1.6 GHzSun UltraSparc IV+ - 1500 MHzI...
Analysis of TPC-H ResultsTPC-H Scale Range of Results10,000 1:2.43,000 1:5.61,000 1:7.4Selecting a suboptimal processorfor...
Analysis of TPC-H ResultsScale really matters!• Itanium• Top @ 10,000• Bottom @ 3,000 and 1,000• Sparc• Top (along with Po...
TPC-H 10,000ResultsProcessorPerf /WeightedCoreQphH CoresCoreFactorIntel Itanium Dual-Core9140 - 1.6 GHz1629 208458 128 1In...
TPC-H 1,000 ResultsProcessorPerf /WeightedCoreQphHCoresCoreFactorIntel QCore Xeon X54503.0GHz45591166976512 0.5Sun SPARC64...
TPC-H 3,000 ResultsProcessorPerf / WeightedCoreQphH CoresCoreFactorIBM POWER5 - 1.9 GHz 2094 100512 64 0.75SPARC64 VII 288...
TPC-H Results DetailProcessorPerf / WeightedCoreQphH CoresCoreFactorIntel Itanium Dual-Core 9140 - 1.6 GHz 1629 208458 128...
Futures•Clock rate will grow•Core weightings will change•Different architecture31
Clock Rate Will GrowCompare architectures.Factor out clock rate, so you canestimate new higher clock ratechips’ performanc...
TPC-C Performance perWeighted Core per GHzProcessorRelative WeightedArchitecture &Design RatingTpmC /WeightedCore / GHzClo...
Top ResultFor TPC-C, Hyper-threadedXeon’s rate highest atPerformance per WeightedCore per GHz.34
Core Weightings WillChangeHypothetically:What would happen if Oracleadopted core weightingsproportional to IBM’sProcessor ...
AnswerFor TPC-C:•Hyper-threaded Xeon’s stay ontop•SPARC drops 24 steps•Power rises 1 step and drops 1,3, 5, & 6 steps•Itan...
TPC-C Ranking Shifts Due toAlternative WeightingsServer CPU TypeOrder due to OracleCore WeightingFactorsOrder due to IBMPr...
Different ArchitectureQ: What will core weightingfactors and PVU’s be for the newAMD Bulldozer processors?(They are somewh...
Upcoming SlideShare
Loading in...5
×

Processor Selection for Optimum Middleware Price Performance

212

Published on

as presented at a poster session at the 2011 Computer Measurement Group Annual Conference

Published in: Technology, News & Politics
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
212
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
2
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Processor Selection for Optimum Middleware Price Performance

  1. 1. Processor Selection ForOptimum MiddlewarePrice / PerformanceDavid A. Kradakra137 at gmail.com1
  2. 2. Which processoris most cost effectiveto run expensivemiddleware for yourapplication?2
  3. 3. One Page SummaryExpensive middleware:• is priced per core differently for different processors;• its costs overshadow server & operating system costs.1. Assess using Performance per Weighted Core(P/WC) rather than Performance per Core.2. The observed best to worst ratio exceeds 7x in one case.3. Results vary not only with middleware and application,but also with the scale.4. For TPC-C on Oracle DBMS, the Hyper-threadedXEONs have the best:• P/WC rating;• Clock-independent architecture:P/WC per GHz.5. For TPC-C on Oracle, similar results come from usingIBM’s PVU rather than Oracle’s Core Weighting Factor.See below forTPC-H and SAPresults as well.3
  4. 4. Processor Selection For Optimum Middleware Price / PerformanceOne Page Summarydakra137 at gmail.comExpensive middleware:• is priced per core differently fordifferent processors, servers,etc;• its costs overshadow server &operating system costs.1. Assess using Performance perWeighted Core (P/WC) ratherthan Performance per Core.2. The observed best to worst ratioexceeds 7x in one case.3. Results vary not only withmiddleware and application, butalso with the scale.4. For TPC-C on Oracle DBMS, theHyper-threaded XEONs have thebest:• P/WC rating;• Clock rate-independentarchitecture:P/WC per GHz.5. For TPC-C on Oracle, similarresults come from using IBM’sPVU rather than Oracle’s CoreWeighting Factor.6. For TPC-H on Oracle:• Itanium• Top @ 10,000• Bottom @ 3,000 and 1,000• Sparc• Top (along with Power) @3,000• 2nd @ 10,000 and 1,000• Power• Top @ 3,000• Hyper-threaded Xeon• Top @ 1,0007. SAP on Oracle• Hyper-threaded Xeon on top8. SAP on DB2• Power7 on top4
  5. 5. AcknowledgementsI wish to thank:• The Transaction Processing Council andSAP for making it so easy to acquirebenchmark results in readily usable form;• Oracle and IBM for making their CoreWeighting Factor and Processor ValueUnit Tables readily available;and especially,• All the staff at all the companies whoimplemented, tuned, measured, andpublished the TPC and SAP benchmarks. 5
  6. 6. Key QuestionHow do you take into account“Core Weighting Factors,”“Processor Value Units,” etc.when looking to pick aprocessor to run EXPEN$IVEmiddleware?6
  7. 7. ExamplesOracle CoreWeightingFactorRatingsAdvantageExampleProcessor0.25 4xOracle SPARCT30.5 2xIntel XEON75xx0.75 1.33x HP PA-RISC1 1x IBM POWER67
  8. 8. AnswerUse“Performance per Weighted Core”rather than“Performance per Core.”8
  9. 9. Performance perWeighted CoreP/WC = P / (Cores x Weight)[where P often is the maximumTPS rate that meets the responsetime requirements.]9
  10. 10. Question #2So what is the most costeffective processor?10
  11. 11. Answer #2aThat really does depend on:• the middleware,• the application, and• its scale.11
  12. 12. Answer #2bThis case studyconsiders:• Middleware:Oracle DBMS• Applications:•TPC-C•TPC-HTPC-H scales:•1,000•3,000•10,000 12
  13. 13. Key PointsMiddleware $ >> Everything else thatvaries e.g. ServerHW & OSThis case study avoided:• Exact pricing, for the reason above• Performing benchmarks• Constraints on publishing results13
  14. 14. TPC-C14
  15. 15. Top 20 TPC-C Results(per ENTERPRISE weighted core)Summary XEON POWER SPARC ItaniumTop 5 TPC-C TPMper weighted core4 1Top 10 TPC-C TPMper weighted core 5 4 1Top 20 TPC-C TPMper weighted core 10 6 1 315
  16. 16. Top 20 TPC-C Results(relative cost per EE weighted core)As of Spring 2011Intel Xeon Processor X5570 2.93GHzIntel Quad-Core Xeon E5520 2.26GHzIntel Xeon E5520 2.27 GHzIBM POWER6 - 4.7 GHzIntel Xeon X5650 6-core 2.66GHzIBM POWER5+ - 2.2 GHzSPARC T3 1.65GHzIntel Xeon Quad-Core X5460 - 3.16 GHzIBM POWER5 - 1.9 GHzIBM POWER5 - 1.9 GHzIBM POWER5 - 1.9 GHzIBM POWER5 - 1.9 GHzIntel Itanium2 Dual-Core - 1.6 GHzIntel Xeon X7460 - 2.67 GHzIntel Xeon QC 5440 - 2.83 GHzIntel Xeon X5355 - 2.66 GhzIntel Xeon X5355 - 2.66 GHzIntel Itanium2 Dual-Core - 1.6 GHzIntel Xeon QC 5440 - 2.83 GHzIntel Itanium2 Dual-Core - 1.6 GHzIntel Itanium Dual-Core 9150M - 1.66 GHzIntel Itanium2 - 1.6 GHz1.0 1.5 2.0 2.5 3.0 3.5 4.01.01.31.41.61.62.02.32.32.32.42.42.62.73.03.03.13.13.13.33.53.73.9Best !!HighestCostsAlmost4x theBest !!16
  17. 17. Key PointDon’t be mislead by the quantities ofbenchmarks!The quantity of entries represents thewillingness of vendors to performbenchmarks and publish results. All itshows is consistency even as otherelements may vary, such as chipset,SAN attachment, storage subsystem, etc.17
  18. 18. Top 20 TPC-C Results(per weighted core) (as of Spring 2011)ProcessorCores /ProcessorCache PerProcessorClockRateCachePerCoreTpmC /WeightedCoreTpmC CoresWeightedCoreFactorIntel Xeon X5570 2.93GHz 4 8 2.93 2 157942 631766 8 0.50Intel QCore Xeon E5520 2.26GHz 4 8 2.26 2 119696 239392 4 0.50Intel Xeon E5520 2.27 GHz 4 8 2.26 2 116001 232002 4 0.50IBM POWER6 - 4.7 GHz 2 8 4.7 4 101116 404462 4 1.00Intel Xeon X5650 6-core 2.66GHz 6 12 2.66 2 96680 290040 6 0.50IBM POWER5+ - 2.2 GHz 78757 236271 4 0.75SPARC T3 1.65GHz 70022 30249688 1728 0.25Intel Xeon QCore X5460 - 3.16GHz 4 12 3.16 3 68417 273666 8 0.50IBM POWER5 - 1.9 GHz 67813 203440 4 0.75IBM POWER5 - 1.9 GHz 66741 1601785 32 0.75IBM POWER5 - 1.9 GHz 64797 194391 4 0.75IBM POWER5 - 1.9 GHz 61841 371044 8 0.75Intel Itanium2 Dual-Core - 1.6 GHz 57642 230569 4 1.00Intel Xeon X7460 - 2.67 GHz 6 16 2.67 2.67 53271 639253 24 0.50Intel Xeon QC 5440 - 2.83 GHz 4 12 2.83 3 52246 104492 4 0.50Intel Xeon X5355 - 2.66 Ghz 4 8 2.66 2 51227 102454 4 0.50Intel Xeon X5355 - 2.66 GHz 4 8 2.66 2 50463 100926 4 0.50Intel Itanium2 Dual-Core - 1.6 GHz 50207 200829 4 1.00Intel Xeon QC 5440 - 2.83 GHz 4 12 2.83 3 48542 97083 4 0.50Intel Itanium2 Dual-Core - 1.6 GHz 44930 359440 8 1.0018
  19. 19. Architecture AnalysisFor TPC-C the Hyper-threaded XEONs come out on topbecause of performance, advantageous core weightingfactor, and clock rate.The Power 6’s are next, due to sheer clock rate, in spite of heaviestweighting.The SPARC T3 would still be in the top 20 chart, ahead of thesome of the XEONs, even if it had the same weighting factor (.5rather than .25) as the XEONs.The best Itanium result compares closely to the Power5 results.While the best XEON rating is 158K, the Itanium results vary from58K down to 45K TpMC/WC rating units.19
  20. 20. Source of Top Xeon’s RatingCache? No, the top three outperformed theXeon’s with larger cache per core and largertotal cache. 2MB per core was enough here.Clock rate? No, the top threeoutperformed the Xeon’s with higher clockrates.Hyper-threading? YES!Core Weighting Factor? YES!20
  21. 21. 21IBM POWER6 - 4.7 GHzIBM POWER5+ - 2.2 GHzSPARC T3 1.65GHzIBM POWER5 - 1.9 GHzIntel Itanium2 Dual-Core - 1.6 GHzIntel Itanium Dual-Core 9150M - 1.66 GHzUltraSPARC T2 Plus 1.6GHzIntel Itanium2 Dual-Core 9050 - 1.6 GHzIBM POWER4 - 1700 MHzIntel Itanium2 - 1.6 GHzIntel Itanium2 - 1.5 GHzFujitsu SPARC64 - 1.3 GHzCHART ABOVE: Standard BELOW: EnterpriseIntel Xeon Processor X5570 2.93GHzIntel Xeon X7460 - 2.67 GHzIntel Xeon X5650 6-core 2.66GHzIntel Xeon Quad-Core X5460 - 3.16 GHzIntel Quad-Core Xeon E5520 2.26GHzIntel Xeon QC 5440 - 2.83 GHzIntel Xeon X5355 - 2.66 GhzIntel Xeon QC 5440 - 2.83 GHzIntel Itanium2 - 1.6 GHz0 2 4 6 8 10 12 14 16 1811.281.441.491.752.392.542.63.163.165.315.431511.982.182.312.646.056.176.5115.67Sections for Standard & Enterprise Editions’ Benchmarks & WeightingsEnterprise Best !!5.4x $Standard Best !!15.7x $(as of Spring 2011)
  22. 22. TPC-H22
  23. 23. TPC-H Results Summary(per weighted core)(as of Spring 2011)Scale10,000Scale 3,000 Scale 1,0001.Itanium(9x40)SPARC (64 VII)& POWER5XEON(x5450)2.SPARC(UltraSPARCIV+)Opteron (285) SPARC 64 VI3. Itanium2ItaniumItanium 23
  24. 24. TPC-H Results Summary(relative cost per weighted core)Intel Itanium Dual-Core 9140 - 1.6 GHzSun UltraSparc IV+ - 1500 MHzIntel Itanium2 - 1.5 GHzSPARC64 VII 2880MHzIntel Itanium2 DC 9050 - 1.6 GHzSun UltraSPARC IV+ - 1800 MHzIntel Itanium2 - 1.5 GHzHP PA-RISC 8700 - 875 MHzSun UltraSPARC III Cu - 1200 MHzIntel Quad-Core Xeon X5450 3.0GHzIntel Itanium Quad-core 9350 - 1.73GHzAMD Dual Core Opteron Model 875HE - 2.2 GHzIntel Itanium2 - 1.6 GHzIntel Itanium2 - 1.6 GHzAMD Opteron Model 848 - 2.2 GHzIntel Itanium2 - 1500 MHz1 2 3 4 5 6 7 811.221.632.122.4211.011.211.771.951.972.152.962.993.713.83.915.6111.852.082.372.462.694.174.234.284.846.237.337.410,000 Best !!3,000 Best !!1,000 Best !!2.4x $5.6x $7.4x $24(as of Spring 2011)
  25. 25. Analysis of TPC-H ResultsTPC-H Scale Range of Results10,000 1:2.43,000 1:5.61,000 1:7.4Selecting a suboptimal processorfor TPC-H could increase costsby up to 7.4x.25
  26. 26. Analysis of TPC-H ResultsScale really matters!• Itanium• Top @ 10,000• Bottom @ 3,000 and 1,000• Sparc• Top (along with Power) @ 3,000• 2nd @ 10,000 and 1,000• Greater gaps between groups of results @ 10,000 and 1,000than @ 3,00026
  27. 27. TPC-H 10,000ResultsProcessorPerf /WeightedCoreQphH CoresCoreFactorIntel Itanium Dual-Core9140 - 1.6 GHz1629 208458 128 1Intel Dual Core Itanium 29040 - 1.6 GHz1339 171380 128 1Sun UltraSparc IV+ - 1500MHz1001 108100 144 0.75Intel Itanium2 - 1.5 GHz 767 49105 64 1Intel Itanium2 - 1.5 GHz674 86283 128 127
  28. 28. TPC-H 1,000 ResultsProcessorPerf /WeightedCoreQphHCoresCoreFactorIntel QCore Xeon X54503.0GHz45591166976512 0.5Sun SPARC64 VI - 2400MHz2470118573640.75Intel Itanium QC 9350 -1.73GHz219014018164 1Intel DC Itanium 91401.6GHz192712332364 1AMD DC Opteron 875HE2.2GHz1855 59354 64 0.528
  29. 29. TPC-H 3,000 ResultsProcessorPerf / WeightedCoreQphH CoresCoreFactorIBM POWER5 - 1.9 GHz 2094 100512 64 0.75SPARC64 VII 2880MHz 2072 198908 128 0.75AMD Opteron DC 285 - 2.6 GHz1728 110577 128 0.5Intel Itanium2 DC 9050 - 1.6 GHz 1182 37814 32 1Fujitsu SPARC64 - 1.3 GHz 1073 34345 64 0.5Sun UltraSPARC IV+ - 1800 MHz1062 114714 144 0.75Sun UltraSparc IV+ - 1500 MHz976 105431 144 0.75Intel Itanium2 - 1.5 GHz 707 45248 64 1Intel Xeon MP - 3.0 GHz 700 22388 32 1HP PA-RISC 8700 - 875 MHz 564 27094 64 0.75Sun UltraSPARC IV - 1200 MHz 550 59436 144 0.75Sun UltraSPARC III Cu - 1200 MHz 536 28948 72 0.75HP PA-RISC 8700 - 750 MHz373 17908 64 0.7529
  30. 30. TPC-H Results DetailProcessorPerf / WeightedCoreQphH CoresCoreFactorIntel Itanium Dual-Core 9140 - 1.6 GHz 1629 208458 1281Intel Dual Core Itanium 2 9040 - 1.6 GHz 1339 171380 1281Sun UltraSparc IV+ - 1500 MHz 1001 108100 1440.75Intel Itanium2 - 1.5 GHz 767 49105 641Intel Itanium2 - 1.5 GHz 674 86283 1281IBM POWER5 - 1.9 GHz 2094 100512 640.75SPARC64 VII 2880MHz 2072 198908 1280.75AMD Opteron Dual-Core Model 285 - 2.6 GHz1728 110577 1280.5Intel Itanium2 DC 9050 - 1.6 GHz 1182 37814 321Fujitsu SPARC64 - 1.3 GHz 1073 34345 640.5Sun UltraSPARC IV+ - 1800 MHz 1062 114714 1440.75Sun UltraSparc IV+ - 1500 MHz 976 105431 1440.75Intel Itanium2 - 1.5 GHz 707 45248 641Intel Xeon MP - 3.0 GHz 700 22388 321HP PA-RISC 8700 - 875 MHz 564 27094 640.75Sun UltraSPARC IV - 1200 MHz 550 59436 1440.7530
  31. 31. Futures•Clock rate will grow•Core weightings will change•Different architecture31
  32. 32. Clock Rate Will GrowCompare architectures.Factor out clock rate, so you canestimate new higher clock ratechips’ performance.Use:Performance per WeightedCore per GHz 32
  33. 33. TPC-C Performance perWeighted Core per GHzProcessorRelative WeightedArchitecture &Design RatingTpmC /WeightedCore / GHzClock(GHz)Perf /WeightedCoreTpmCXeon X5570 2.93GHz 1.00 53905 2.93 157942 631766SPARC T3 1.65GHz 0.79 42438 1.65 70022 30249688Xeon X5650 6C 2.66GHz 0.67 36346 2.66 96680 290040Itanium2 DC 1.6 GHz 0.67 36026 1.6 57642 230569POWER5+ 2.2 GHz 0.66 35799 2.2 78757 236271POWER5 1.9 GHz 0.66 35691 1.9 67813 203440Itanium2 DC - 1.6 GHz 0.58 31380 1.6 50207 200829Xeon QC X5460 - 3.16 GHz 0.40 21651 3.16 68417 273666POWER6 4.7 GHz 0.40 21514 4.7 101116 404462Xeon X7460 - 2.67 GHz0.37 19952 2.67 53271 639253Similar rows omitted 33
  34. 34. Top ResultFor TPC-C, Hyper-threadedXeon’s rate highest atPerformance per WeightedCore per GHz.34
  35. 35. Core Weightings WillChangeHypothetically:What would happen if Oracleadopted core weightingsproportional to IBM’sProcessor Value Units (PVU’s)?35
  36. 36. AnswerFor TPC-C:•Hyper-threaded Xeon’s stay ontop•SPARC drops 24 steps•Power rises 1 step and drops 1,3, 5, & 6 steps•Itanium rises 4 & 5 steps. 36
  37. 37. TPC-C Ranking Shifts Due toAlternative WeightingsServer CPU TypeOrder due to OracleCore WeightingFactorsOrder due to IBMProcessor ValueUnitsOrderDifferenceIntel Xeon Processor X5570 2.93GHz 1 1 0Intel Xeon E5520 2.27 GHz 3 4 -1IBM POWER6 - 4.7 GHz 4 3 1Intel Xeon X5650 6-core 2.66GHz 5 5 0IBM POWER5+ - 2.2 GHz 6 7 -1SPARC T3 1.65GHz 7 31 -24Intel Xeon Quad-Core X5460 - 3.16 GHz 8 6 2IBM POWER5 - 1.9 GHz 9 12 -3IBM POWER5 - 1.9 GHz 10 15 -5IBM POWER5 - 1.9 GHz 12 18 -6Intel Itanium2 Dual-Core - 1.6 GHz 13 8 5Intel Xeon X7460 - 2.67 GHz 14 9 5Intel Xeon X5355 - 2.66 GHz 17 13 4Intel Itanium2 Dual-Core - 1.6 GHz 18 14 4Intel Xeon QC 5440 - 2.83 GHz 19 17 2Similar rows omitted 37
  38. 38. Different ArchitectureQ: What will core weightingfactors and PVU’s be for the newAMD Bulldozer processors?(They are somewhat less than 2 cores perprocessor building block.)A: I don’t know.38
  1. A particular slide catching your eye?

    Clipping is a handy way to collect important slides you want to go back to later.

×