INFOBrief Dell - Your next ®SAP NetWeaver Portal-based Solution? Results • On April 7, 2006, SAP AG certified the results of a two‐tier SAP® Employee Self‐Service Portal (EP‐ESS) Standard Application Benchmark on a Dell PowerEdge 6850 server reaching a total of 1941 1SAP EP‐ESS Benchmark users . • Dell demonstrates its technology leadership in delivering the highest 2two‐tier SAP EP‐ESS Benchmark result as of May 4, 2006. • The purpose of this INFOBrief is to share Dellʹs lessons learned in setting up a stable, long‐lasting SAP NetWeaver® Portal‐based solution. Main steps that may assist in increasing performance and thus help to lower TCO can be found in this document. Dell Product Group/Enterprise Page 1 April 2006 UPUP The Challenge and the Configuration As the enterprise services architecture (ESA) gains momentum, more and more customers take the chance to integrate the SAP® NetWeaver Portal as a collaboration platform into their SAP NetWeaver platform‐based landscape. Because the Java‐based portal is quite different from ABAP‐based SAP® applications such as Sales and Distribution or Materials Management, research into optimal configuration settings was needed. The benchmark environment used is based on Novellʹs SUSE Linux Enterprise Server. Many customers with longstanding UNIX® backgrounds and investments are looking to Linux as a more cost‐effective alternative ...
The Challenge and the Configuration U Astheenterpriseservicesarchitecture(ESA)gainsmomentum,moreandmorecustomerstakethechancetointegratetheSAP®NetWeaverPortalasacollaborationplatformintotheirSAPNetWeaverplatform‐basedlandscape.BecausetheJava‐basedportalisquitedifferentfromABAP‐ basedSAP®applicationssuchasSalesandDistributionorMaterialsManagement,researchintooptimalconfigurationsettingswasneeded.ThebenchmarkenvironmentusedisbasedonNovellʹsSUSELinuxEnterpriseServer.ManycustomerswithlongstandingUNIX®backgroundsandinvestmentsarelookingtoLinuxasamorecost‐effectivealternative.BenchmarkdetailsU Thebenchmark’sconfigurationandresultsare:
Hardware
Application Database Operating System SAP EPESS Benchmark users
Number of Top Level Navigation steps/hour
Dell Product Group/Enterprise
Dell™PowerEdge™68504processors/8cores/16threadsIntelXeonMP3.0Ghz32 GB RAM SAP NetWeaver Portal 6.0 MaxDB 7.5 3 Novell® SUSE™ Linux Enterprise Server 9 EM64T P P
1,941 583 000
Page 2
April 2006
What does it mean for your business? TheSAPEP‐ESSStandardApplicationBenchmarksimulatesreal‐lifeusersdoingtypicalemployeeactivitieslikerecordingtheirworkingtimeorlookingattheirbankinformation.Theback‐endinfluenceontheperformancehasbeenomitted.Thegoalistoreachanaverageresponsetimeoflessthantwosecondswithamaximumusernumber.Withthisusernumberinmind,youcancomparedifferenthardwareplatformsfortheneedsofyourbusinessplatform.Performance Parameters TherearemanyconfigurationparametersfortheSAP®NetWeaverPortalthatcanbefine‐tunedforperformance.Thechallengingtaskistofindthosethatreallymakeadifferencewithregardstobusinessneeds.Forexample,withUNIX/Linuxsystems,thereisoftenthenotionthatrunningagraphicinterfacecouldadverselyaffecttheoverallsystemperformance,soyoucouldgainperformancebydisablingit.Toverifythis,DellranthebenchmarkoncewithafullyfeaturedKDEdesktopenvironmentrunning,andoncewithout.TherunwiththeKDEgraphicinterfaceshowednosignificantchangesintheresponsetime.Thisshowsthat‐providedtheabsenceofresourcebottlenecks‐theactivationofasystemʹsdesktopinterfacewouldnottypicallyinterferewithitsportalsfunction.Thefollowingaresettingsthatthisbenchmarkshowedtobeofsignificance.BIOSSettingsThehardwareprefetcheroptimizesperformanceforsequentialmemoryaccess.Wefoundanincreaseinperformanceifitisdisabled,whilethePowerEdge6850’sdefaultsettingshaveitenabled.Toturnitoff,dothefollowing:−Whenbootingtheserver,entertheDellsystemBIOSbypressingF2−InthesystemBIOS,choseCPUinformation‐>SequentialMemoryAccess−Setthisoptionto“disabled”WefoundthattheprefetcherhadthefollowingperformanceimpactontheSAPNetWeaverPortal‐basedsystem:
Dell Product Group/Enterprise
Page 3
April 2006
9
8 7 6
5 4
3 2 1
0
Prefetcher's influence
response time
OFF ON state Figure 1 Performance impact of thePrefetchersetting (lower is better) Figure1showsthatswitchingofftheprefetcherdecreased,inthisparticularexample,thesystemʹsansweringtimefrom8.47secondsto2.16seconds,whileallotherparametersremainedunchanged.
0 6 7 8 server nodes Figure 3 Performance impact of the server nodes' count(lower is better)Thediagraminfigurethreeshowstheminimumresponsetimeoftheexamplesystematsevenservernodeswith1.92seconds,comparedto2.16secondsforsixandeightservernodes.Foreachcolumn,thethreadcountforallservernodesremainedunchanged.
0 30 40 50 threads Figure 4 Performance impact of the threads' number(lower is better)Thediagraminfigurefourshowstheminimumresponsetimeoftheexamplesystemat40threadsperservernode.Theresponsetimeis1.92secondshere,comparedto1.94at30and2.02at50threadspernode.Allotherparametersremainedunchanged.Bottom Line U TuningtherightparametersinanSAP®NetWeaverPortal‐basedsystemdeliverssignificantperformanceincreases.Whileinourteststhehardware(BIOS)settingshaveyieldedthemostimportanteffect,thesoftwareparametersplayanimportantroleinfinetuningthesystem,especiallywhenmappingthesoftwareconfigurationtothehardwareresourcesavailable.DistinctCPUormemoryresourcesneedadaptedsettingsinordertobeoptimallyused.Findingtherightfithelpsyourbusinessplatformtoruninamorestablemanner,deliveringthehighestpossibleperformanceandthusprovidingamoreoptimalreturnoninvestment.