《Leveling the Playing Field Industry Standard Benchmarks and OCP.pdf》由会员分享,可在线阅读,更多相关《Leveling the Playing Field Industry Standard Benchmarks and OCP.pdf(15页珍藏版)》请在三个皮匠报告上搜索。
1、A discussion about the role of benchmarking in impending legislation and how OCP is treated differently to OEM hardwareLeveling the Playing Field:Industry Standard Benchmarks and OCPDr Kat Burdett,Research and Development Engineer,TechbuyerLeveling the Playing Field:Industry Standard Benchmarks and
2、OCPDC SUSTAINABILITYSUSTAINABILITYThe growing Data Centre industry has caught the attention of various legislative bodies around the worldSome countries or municipalities have even opted for moratoriums on new data centresThe building of new and growth of existing data centres in those regions will
3、require participation in energy efficiency directivesVarious directives(e.g.,EnergyStar in the US and Energy Efficiency Directive&EcoDesign Directive in the EU)mandate reporting servers energy efficiency based on SPEC SERT 2.0 benchmark.SERT is also part of the accepted benchmarks for the ITEEsv KPI
4、(ISO/IEC 30134-4)BackgroundDesigned for measuring and evaluating the energy efficiency of servers Separate workloads for CPU,Memory and Storage plus Idle:7 CPU,2 Memory and 2 Stroage Worklets test at multiple load levelsSERT MetricSPEC SERT 2.0 SuiteCPU 65%(High)Memory 30%(Medium)Storage 5%(Low)Work
5、load WeightingSERT2.0 Benchmarking SetupSUT Configuration:General availability of SUT with deviations from defaults documentedMemory,Network,storage,JVM configuration rulesBenchmark configuration based on defined CPU,OS,JVM tuples Electricity Source:Preferred AC by local utility companiesInternal ba
6、tteries must be fully chargedDC Line voltages not currently supported or testedTest Setup:Limited range of power analyzers supported by PTDaemon No DC power analyzer approved No active components between the SUT and power analyzerRun and Reporting RulesQuantifying energy efficiency for data centre h
7、ardware is becoming a requirement for operation or expansionThe OCP community faces challenges in meeting standardized energy efficiency metrics compared to OEMsCurrently not possible to run SPEC SERT2.0 benchmark on OCP hardware for reasons including but not limited to:Very limited DC power analyze
8、r supportStrict power monitoring methodology not compatible with powershelf configurationA qualitive prerequisite for availability in the marketplaceProblem StatementCollaborative approach:a partnership between RISE and TechbuyerServer Testing Wind-tunnelWe use it to control the temperature,pressure
9、 and airflow to profile different servers and mimic their behaviour in a real data centre environmentWe use it to profile different makes,models,and generations of server and their different configurationsThis allows us to develop the relationship between environment,energy efficiency,and performanc
10、e for a serverServer Testing Wind-tunnelSERT2.0 prohibits the placing of hardware between the power analyzer and the server,such as a powershelfSERT2.0 does,however,support the benchmarking of blade servers in blade enclosuresWhile the analogy does not map perfectly,we propose building on this simil
11、arity in configuration to create OCP/OEM testing parityProposed MethodologyORV2 vs Blade EnclosureHP BLc7000 G3 Open Rack V2SERT2.0 prohibits the placing of hardware between the power analyzer and the server,such as a powershelfSERT2.0 does,however,support the benchmarking of blade servers in blade
12、enclosuresWhile the analogy does not map perfectly,we propose building on this similarity in configuration to create OCP/OEM testing parityFurthermore,we should be able to isolate sled performance from any powershelf overhead in the same way we have isolated the behavior of blades in enclosures by m
13、odulating number of installed nodes during testingProposed MethodologyHelp bring OCP into the mainstreamBring OCP into compliance with approaching regulationsAssistance from OCP DC Efficiency Metrics working group would be invaluableIncrease use of OCP hardware for second-hand and enterprise marketsProvide quantifiable comparison for OCP vs OEMSupply OCP hardware for testing to broaden the approach for parity methodologyContact us if you think you can helpresearch-Thank you!