sanitysisoftwaresandra

TESTREPORTFEBRUARY2006SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversTESTREPORTDECEMBER2007KEYFINDINGSzTheDellPowerEdgeM600achievedbetterperformance/wattthantheHPBladeSystemc-ClassortheIBMBladeCenterHType8852ateveryconfigurationwetested(seeFigure1).
zWith10bladesinstalledinallthreesystems,thebladesintheDellPowerEdgeM600achieved24.
59percenthigherperformance/wattthantheHPBladeSystemc-Classand28.
76percenthigherthantheIBMBladeCenterHType8852.
zWith16bladesineachchassis,theDellPowerEdgeM600achieved25.
35percentbetterperformance/wattthantheHPBladeSystemc-Class.
zAtthemaximumbladesconfiguration,theDellPowerEdgeM600achieved28.
58percentbetterperformance/wattwith16bladesthantheIBMBladeCenterHType8852didwith14blades.
zWith16blades,theDellPowerEdgeM600used18.
55percentlesspowerperbladethantheHPBladeSystemc-Classdidwith16bladesand11.
76percentlesspowerperbladethantheIBMBladeCenterHType8852didwith14blades.
ExecutivesummaryDellInc.
(Dell)commissionedPrincipledTechnologies(PT)tomeasuretheSPECjbb2005performanceperwattofthefollowingdual-socketbladeservers:zDellPowerEdgeM600zHPBladeSystemc-ClasszIBMBladeCenterHType8852Inthissection,wepresentthebestresultsforeachserver.
ForcompletedetailsoftheperformanceofeachJavaVirtualMachine(JVM)bywarehouseforeachserver,seetheTestresultssection.
Figure1presentstheperformance/wattforeachofthebladeserversystemsbyconfiguration.
Higherresultsshowbettersystemperformance/watt.
Wecalculateperformance/wattbydividingtheSPECjbb2005resultbytheaveragepowerconsumptioninwattsduringtheperiodthesystemachievedpeakperformance.
Figure1showsthebladesintheDellPowerEdgeM600achievedthebestperformance/wattateveryconfiguration.
With10bladesinstalledinallthreesystems,theDellPowerEdgeM600achieved24.
59percenthigherperformance/wattthantheHPBladeSystemc-Class.
TheDellPowerEdgeM600alsoachieved28.
76percenthigherperformance/wattthantheIBMBladeCenterHType8852.
Inthemaximumbladesconfiguration,weinstalledthemaximumnumberofbladeserversineachchassis.
Thiswas16serversfortheDellandHP,butonly14bladesfortheIBM.
Inthemaximumbladesconfiguration,TheDellPowerEdgeM600achieved25.
35percentbetterperformance/wattwith16bladesinstalledthantheHPBladeSystemc-Classdidwith16bladesinstalled.
TheDellPowerEdgeM600achieved28.
58percentbetterperformance/wattwith16bladesinstalledthantheIBMBladeCenterHType8852with14bladesinstalled.
IncomparingthemaximumbladesconfigurationfortheDellPowerEdgeM600andtheIBMBladeCenterHType8852,weusedthemaximumnumberofbladeseachserverwouldsupport:16intheDelland14intheIBM.
Inthiscase,wearethereforenormalizinga16-bladeresulttoa14-bladeresult.
Furthermore,theDellPowerEdgeM600achieved31.
95and27.
83percenthigherperformance/wattthantheHPBladeSystemc-Classinone-andtwo-bladeconfigurations,respectively.
TheDellPowerEdgeM600achieved88.
16and65.
17percenthigherperformance/wattthantheIBMBladeCenterHType8852intheone-andtwo-bladeconfigurations,respectively.
2PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversDellPowerEdgeM600BladeSystemHPBladeSystemc-ClassIBMBladeCenterHType8852Percentageperformance/wattincreaseDelloverHPPercentageperformance/wattincreaseDelloverIBM1blade464.
54352.
06246.
8931.
9588.
162blades642.
40502.
52388.
9327.
8365.
1710blades919.
95738.
40714.
4724.
5928.
76Maximumblades958.
86(16blades)764.
97(16blades)745.
70(14blades)25.
3528.
58Figure1:Performance/wattresultsforeachserverbybladeconfiguration.
Highernumbersarebetter.
WorkloadSPECjbb2005isanindustry-standardbenchmarkcreatedbytheStandardPerformanceEvaluationCorp.
(SPEC)tomeasureaserver'sJavaperformance.
SPECmodeledSPECjbb2005onthethree-tierclient/serverarchitecture,withthemiddlelayerastheprimaryfocus.
AccordingtoSPEC,"Randominputselectionrepresentsthefirst(user)tier.
SPECjbb2005fullyimplementsthemiddletierbusinesslogic.
Thethirdtierisrepresentedbytablesofobjects,implementedbyJavaCollections,ratherthanaseparatedatabase.
"(www.
spec.
org/jbb2005/docs/UserGuide.
html).
SPECjbb2005utilizesmultiplespecialdatagroupsandmultiplethreadsasitruns.
Eachdataunitisa"warehouse,"whichisaroughly25MBcollectionofdataobjects.
Eachthreadrepresentsanactiveuserpostingtransactionrequestswithinawarehouse.
Thebenchmarkrunbeginswithonewarehouseandthenincreasesthenumberofwarehouses;itsgoalistosaturatetheserver'sprocessorcapacity.
Asthenumberofwarehousesincreases,sodoesthenumberofthreads.
Thebenchmark'sresultsportraytheserver'sthroughputinbusinessoperationspersecondorSPECjbb2005bops.
AhighernumberofSPECjbb2005bopsisbetter.
(FormoreinformationonSPECjbb2005,gotowww.
spec.
org.
)TestresultsFortesting,weinstalledagivennumberofbladeserversintothechassisandranSPECjbb2005onallservers.
BeforestartingtheSPECjbb2005benchmark,weloggedintothesystemandallowedtheserverstositidlefor10minutes.
Wethenstartedrecordingpowerfor2minutes.
Thisprocessmeantthatallsystemswereidlefor12minutesbeforewebeganthebenchmark.
Ineachtestconfiguration,weran2JVMinstancesatthesametime,acommonpracticeonserverswithmanyprocessors.
Tocomputetheoverallscoreforthesystem,SPECjbb2005sumsthescoresofalltheJVMs.
SPECjbb2005computesthescoreofeachJVMbytakingtheaverageoftheresultsduringmixeswhentheserverisrunningatpeakperformance.
Inourtesting,allserversachievedpeakperformanceduringmixes4through8.
(InSPEC'sterms,theseresultsarefrom"compliant"runs,meaningthatwecandisclosethempubliclywithoutpostingthemontheSPECWebsitewithallthefilesSPECusuallyrequires.
Wedopresenthereallthedatanecessarytoreproducetheseresults.
)Inthetablesbelow,weshowtheSPECjbb2005resultsforeachbladeforagivenconfiguration.
Figure2showstheidlepowerusage(inwatts)fortheDellPowerEdgeM600,HPBladeSystemc-Class,andIBMBladeCenterHType8852atallbladeconfigurations.
Lowerpowerisbetter.
3PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversDellPowerEdgeM600HPBladeSystemc-ClassIBMBladeCenterHType88521blade383.
75510.
57681.
882blades516.
68666.
00807.
3210blades1,588.
151,975.
381,853.
73Maximumblades2,416.
18(16blades)3,035.
87(16blades)2,409.
92(14blades)Figure2:Idlepowerusage(inwatts)ofthetestserversduringthemedianpeakrunsforeachbladeconfiguration.
Lowernumbersarebetter.
Figure3showstheaveragepowerusage(inwatts)forDellPowerEdgeM600,HPBladeSystemc-Class,andIBMBladeCenterHType8852atallbladeconfigurations.
Lowerpowerisbetter.
Tocalculatetheaveragepower,werecordedthepowerduringtheSPECjbb2005benchmarkandaveragedthepowerduringtheperiodthesystemachievedpeakperformance.
ThepowertheDellPowerEdgeM600requiredrunningwith16bladesisalmostidenticaltothepowertheIBMBladeCenterHType8853requiredrunningwith14blades.
DellPowerEdgeM600HPBladeSystemc-ClassIBMBladeCenterHType88521blade454.
39590.
01749.
572blades656.
45821.
71954.
7910blades2,277.
472,802.
432,605.
88Maximumblades3,524.
19(16blades)4,326.
92(16blades)3,494.
45(14blades)Figure3:Averagepowerusage(inwatts)ofthetestserversduringthemedianpeakrunsforeachbladeconfiguration.
Lowernumbersarebetter.
Tocalculatetheperformance/wattweusedthefollowingformula:Performance/watt=thebenchmark'sscore/averagepowerconsumptioninwattsduringtheperiodthesystemachievedpeakperformance.
Forthe2-,10-,maximum-bladeconfigurations,wedividedtheaveragepowershowninFigure3bythenumberofblades.
Wethendividedthebenchmark'sscoreforeachbladebytheaveragepower.
Theformulaforthesebladeconfigurationswereasfollows:Performance/wattbyblade(2-,10-,maximum-bladeconfigurations)=(benchmarkscorebyblade/[totalaveragepower/numberofblades])Wethenaveragedtheperformance/wattforallbladesinthegivenconfigurations.
Foreachconfiguration,weperformed3runsofSPECjbb2005andrecordedthepowerduringtheseruns.
Theresultsbelowarethemedianof3testruns.
Figure4showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheDellPowerEdgeM600with1bladeinstalledinthechassis.
SPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1211,081454.
39464.
54Figure4:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheDellPowerEdgeM600duringthemedianrunforthe1-bladeconfiguration.
Higherperformance/wattisbetter.
4PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversFigure5showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheDellPowerEdgeM600with2bladesinstalledinthechassis.
Wecalculatedtheaverageperformance/wattbyaveragingtheperformance/wattscoresofthe2systems.
SPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1210,754328.
22642.
11System2210,946328.
22642.
69Averageperformance/watt642.
40Figure5:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheDellPowerEdgeM600duringthemedianrunforthe2-bladeconfiguration.
Higherperformance/wattisbetter.
Figure6showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheDellPowerEdgeM600with10bladesinstalledinthechassis.
Wecalculatedtheaverageperformance/wattbyaveragingtheperformance/wattscoresofthe10systems.
SPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1209,976227.
75921.
97System2211,440227.
75928.
40System3212,122227.
75931.
39System4196,571227.
75863.
11System5211,368227.
75928.
08System6211,277227.
75927.
68System7209,919227.
75921.
72System8210,591227.
75924.
67System9211,792227.
75929.
94System10210,096227.
75922.
50Averageperformance/watt919.
95Figure6:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheDellPowerEdgeM600duringthemedianrunforthe10-bladeconfiguration.
Higherperformance/wattisbetter.
Figure7showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheDellPowerEdgeM600with16bladesinstalledinthechassis.
Wecalculatedtheaverageperformance/wattbyaveragingtheperformance/wattscoresofthe16systems.
5PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversSPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1211,626220.
26960.
79System2210,538220.
26955.
85System3211,956220.
26962.
29System4211,223220.
26958.
96System5211,357220.
26959.
57System6210,810220.
26957.
09System7210,810220.
26957.
09System8210,289220.
26954.
72System9210,827220.
26957.
17System10210,982220.
26957.
87System11211,456220.
26960.
02System12210,711220.
26956.
64System13211,745220.
26961.
33System14211,525220.
26960.
34System15212,135220.
26963.
10System16211,218220.
26958.
94Averageperformance/watt958.
86Figure7:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheDellPowerEdgeM600duringthemedianrunforthe16-bladeconfiguration.
Higherperformance/wattisbetter.
Figure8showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheHPBladeSystemc-Classwith1bladeinstalledinthechassis.
SPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1207,723590.
01352.
06Figure8:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheHPBladeSystemc-Classduringthemedianrunforthe1-bladeconfiguration.
Higherperformance/wattisbetter.
Figure9showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheHPBladeSystemc-Classwith2bladesinstalledinthechassis.
Wecalculatedtheaverageperformance/wattbyaveragingtheperformance/wattscoresofthe2systems.
SPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1206,882410.
86503.
54System2206,048410.
86501.
51Averageperformance/watt502.
52Figure9:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheHPBladeSystemc-Classduringthemedianrunforthe2-bladeconfiguration.
Higherperformance/wattisbetter.
Figure10showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheHPBladeSystemc-Classwith10bladesinstalledinthechassis.
Wecalculatedtheaverageperformance/wattbyaveragingtheperformance/wattscoresofthe10systems.
6PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversSPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1206,358280.
24736.
35System2206,638280.
24737.
35System3206,655280.
24737.
41System4206,708280.
24737.
60System5207,438280.
24740.
21System6206,919280.
24738.
36System7206,860280.
24738.
14System8207,330280.
24739.
82System9206,647280.
24737.
38System10207,770280.
24741.
39Averageperformance/watt738.
40Figure10:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheHPBladeSystemc-Classduringthemedianrunforthe10-bladeconfiguration.
Higherperformance/wattisbetter.
Figure11showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheHPBladeSystemc-Classwith16bladesinstalledinthechassis.
Wecalculatedtheaverageperformance/wattbyaveragingtheperformance/wattscoresofthe16systems.
SPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1207,182270.
43766.
11System2207,103270.
43765.
82System3206,311270.
43762.
89System4208,405270.
43770.
64System5206,547270.
43763.
77System6206,787270.
43764.
65System7207,264270.
43766.
42System8208,406270.
43770.
64System9205,203270.
43758.
80System10207,253270.
43766.
38System11206,343270.
43763.
01System12206,371270.
43763.
11System13206,892270.
43765.
04System14207,169270.
43766.
07System15206,270270.
43762.
74System16206,467270.
43763.
47Averageperformance/watt764.
97Figure11:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheHPBladeSystemc-Classduringthemedianrunforthe16-bladeconfiguration.
Higherperformance/wattisbetter.
Figure12showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheIBMBladeCenterHType8852with1bladeinstalledinthechassis.
7PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversSPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1185,163749.
76246.
96Figure12:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheIBMBladeCenterHType8852duringthemedianrunforthe1-bladeconfiguration.
Higherperformance/wattisbetter.
Figure13showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheIBMBladeCenterHType8852with2bladesinstalledinthechassis.
Wecalculatedtheaverageperformance/wattbyaveragingtheperformance/wattscoresofthe2systems.
SPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1185,080477.
39387.
69System2186,269477.
39390.
18Averageperformance/watt388.
93Figure13:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheIBMBladeCenterHType8852duringthemedianrunforthe2-bladeconfiguration.
Higherperformance/wattisbetter.
Figure14showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheIBMBladeCenterTypeH8852with10bladesinstalledinthechassis.
Wecalculatedtheaverageperformance/wattbyaveragingtheperformance/wattscoresofthe10systems.
SPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1187,011260.
59717.
65System2185,789260.
59712.
96System3186,519260.
59715.
76System4186,310260.
59714.
96System5186,038260.
59713.
92System6186,163260.
59714.
40System7185,962260.
59713.
62System8186,117260.
59714.
22System9186,921260.
59717.
30System10184,995260.
59709.
91Averageperformance/watt714.
47Figure14:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheIBMBladeCenterHType8852duringthemedianrunforthe10-bladeconfiguration.
Higherperformance/wattisbetter.
Figure15showstheSPECjbb2005results,averagepowerperblade,andperformance/wattfortheIBMBladeCenterHType8852with14bladesinstalledinthechassis.
Wecalculatedtheaverageperformance/wattbyaveragingtheperformance/wattscoresofthe14systems.
8PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversSPECjbb2005bopsAveragepowerperblade(watts)Performance/wattSystem1186,849249.
60748.
58System2184,574249.
60739.
47System3186,015249.
60745.
24System4186,826249.
60748.
49System5185,888249.
60744.
73System6185,400249.
60742.
78System7186,443249.
60746.
96System8186,996249.
60749.
17System9186,388249.
60746.
74System10186,488249.
60747.
14System11186,060249.
60745.
42System12186,037249.
60745.
33System13186,044249.
60745.
36System14185,818249.
60744.
45Averageperformance/watt745.
70Figure15:SPECjbb2005results,averagepowerusage(inwatts),andperformance/wattfortheIBMBladeCenterHType8852duringthemedianrunforthe14-bladeconfiguration.
Higherperformance/wattisbetter.
ConcernedaboutIBM'srelativelylowSPECjbbscores,weinvestigatedfurther.
Weverifiedthatthesystem'sprocessorswererunningatfullspeedandtherewerenoheatissues.
Wetestedallsystemsusing4GBofRAM.
TheIBMBladeCenterHType8852hadonly4RAMslots,allofwhichwerefull.
TheDellPowerEdgeandHPBladeSystemeachhad8RAMslots,ofwhichweused4.
RAMwasrunningat667MHzforallthreetestsystems.
Asasanitycheck,weranSiSoftwareSandra'smemorytest,whichreportedtheIBMserver'smemorybandwidthashalfthatoftheDellserver's:10,656MB/svs.
21,312MB/s.
SiSoftwareSandraalsoreportedtheIBMserverhad2memorychannels,whiletheDellserverhad4memorychannels.
SandrashowedonlythattheIBMBladeCenterHType8852usedanIntel5000Pchipset,withnocodenamesorotherdetails.
Itdidreportthatthememorycontrollersupports8memoryslots,thesameastheDellandtheHP.
ItispossiblethatusingtheoptionalIBMBladeCenterMemoryandI/OExpansionBlade,whichprovides4additionalDIMMslots,isnecessarytoaccesstheothertwomemorychannels.
TheIBMdocumentationsaysthatwhentheusingtheMemoryexpansionblade,theoptimalmemoryconfigurationistouse2pairsofmatchingDIMMs;1pairontheboardinslots1and3andtheotherpairontheexpansionbladeinslots5and7.
Becausethesystemwetesteddidnotshipwiththispart,wewereunabletotestthishypothesis.
Inconclusion,webelievethattheIBMBladeCenterHType8852'srelativelylowSPECjbbscoresareprobablyafunctionofitslowermemorybandwidth.
PowersettingsexperimentsWeranonetestwiththemaximumnumberofbladesforeachenclosureandthepowersettingsontheenclosuresasidenticalaspossible.
Weused"ACredundant"fortheDellPowerEdgeM600,"ACredundant"fortheHPBladeSystemc-Class,and"Redundantwithoutperformanceimpact"fortheIBMBladeCenterHType8852.
Figure16showstheresultsofthattest.
Asyoucansee,theresultsweresoclosetothoseofourmainteststhatwefocusedontheresultsofthosetests.
(Aswenotedearlier,themaximumnumberofbladesis16fortheDellPowerEdgeM600andHPBladeSystemc-Class,butonly14fortheIBMBladeCenterHType8852.
)9PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversDellPowerEdgeM600HPBladeSystemc-ClassIBMBladeCenterHType8852Averageperformanceperwatt(higherisbetter)949.
36764.
97762.
99Averageidlepower,watts(lowerisbetter)2413.
853,035.
872392.
73Averagerunpowerperblade,watts(lowerisbetter)222.
31270.
43243.
94Averageperformanceperblade,BOPS(higherisbetter)211,054(16blades)206,873.
31(16blades)186,121.
79(14blades)Figure16:Resultsofsingletestsusingeachenclosure'smaximumnumberofbladesandthefollowingpowersettings:"ACredundant"fortheDellPowerEdgeM600,"ACredundant"fortheHPBladeSystemc-Class,and"Redundantwithoutperformanceimpact"fortheIBMBladeCenterHType8852.
TestmethodologyPerformthefollowingBIOS-leveloperationsoneverysystembeforeinstallingtheOS:1.
MakesuresystemsareconfiguredwithRAID1.
Usethediskcontrollerutilityforthis,nottheOS.
2.
Setthepartitiontobethewholedisk.
3.
UpdatetheBIOS,ifneeded.
Note:ThepreinstalledBIOS,version1.
05,ontheIBMBladeCenterHType8852gavetheerror"TheBIOSdoesnotsupportthecurrentsteppingofProcessorP02".
Afterupgradingtoversion1.
07,theerrormessagenolongerappeared.
4.
DisableHWprefetcherandAdjacentlineprefetcherinBIOS.
Leaveallothervaluesattheirdefaults.
Createthebaseimage:1.
Foreachinstallation,beginbyinstallingafreshcopyofMicrosoftWindows2003ServerEnterprisex64EditionServicePack2oneachbladeserver.
(Ifyouareusingasupportpackwithanexpressinstalloption,donotuseit.
SelectCustomInstallandonlyinstalldrivers.
Otherwise,youmaybeinstallingunnecessarysoftware,whichmayaffecttheresultsofthetest.
)a.
Forthelicensingmode,usethedefaultsettingoffiveconcurrentconnections.
b.
Enterapasswordfortheadministratorlogon.
c.
SelectEasternTimeZone,andcheckdateandtime.
d.
UsetypicalsettingsfortheNetworkinstallation.
e.
Assignacomputername.
Weusedtheformat"Server",whereBrandisDell,HP,orIBMandXisthebladenumberinthechassis(1–16forDellandHPand1–14forIBM)(e.
g.
,IBMServer1).
f.
Leavethedefault"WORKGROUP"fortheworkgroup.
g.
Finishinstallation.
h.
InstallSP2.
i.
RunLiveupdateandinstallthefollowingupdates.
OurupdatedatewasNovember29,2007.
WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB943460)WindowsServer2003WindowsMaliciousSoftwareRemovalToolx64-November2007(KB890830)WindowsServer2003CumulativeSecurityUpdateforInternetExplorer6forWindowsServer2003x64Edition(KB939653)WindowsServer2003SecurityUpdateforOutlookExpressforWindowsServer2003x64Edition(KB941202)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB933729)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB936021)WindowsServer2003UpdateforWindowsServer2003x64Edition(KB933360)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB938127)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB921503)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB936782)10PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversWindowsServer2003UpdateforWindowsServer2003x64Edition(KB932596)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB926122)WindowsServer2003SecurityUpdateforWindowsMediaPlayer6.
4(KB925398)WindowsServer2003UpdateforWindowsServer2003x64Edition(KB936357)WindowsServer2003CumulativeSecurityUpdateforOutlookExpressforWindowsServer2003x64Edition(KB929123)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB935839)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB935840)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB924667)WindowsServer2003UpdateforWindowsServer2003x64Edition(KB927891)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB932168)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB930178)WindowsServer2003SecurityUpdateforWindowsServer2003x64Edition(KB925902)2.
ToimproveJavaperformance,enablelargepagesinmemoryonallservers.
Toenablethisservice,theadministratormustfirstassignadditionalprivilegestotheuserwhowillberunningtheapplication.
(Weassignedthisprivilegeonlytotheadministrator,becauseweusedthataccountforourtests.
)Toenablelargepages,selectthefollowing:ControlPanelAdministrativeToolsLocalSecurityPolicyLocalPoliciesUserRightsAssignment:AddAdministrator"Lockpagesinmemory,"addusersand/orgroups3.
Turnthescreensaveroff.
4.
CustomizedesktoptoputMyComputeronit.
5.
Setscreento10x7,32-bitcolor.
6.
IntheManageyourserverdialog,select"Don'tdisplaythispageatlogon.
"7.
TurnoffAutomaticupdates.
8.
Settheservertologinautomatically.
9.
InstallSPECjbbandtheJVMontheserver10.
Emptytrash.
11.
Createtheimageonserver.
Forallotherblades,useGhostcasttoinstalltheimage.
UsethesystemwejustcompletedconfiguringtocreateaGhostimageontheserver.
Aftercapturingimagestoservers1.
Eachserverneedsauniquehostname.
Changethehostnamefromtheoneyouusedforthebaseimagetooneoftheform"Server",whereBrandisDell,HP,orIBMandXisthebladenumberinthechassis(1–16forDellandHPand1–14forIBM)(e.
g.
,IBMServer2).
Wedidasmallamountofexperimentingwiththepowersettingsontheenclosures.
FortheDellPowerEdgeM600,wetestedwiththeRedundancyPolicytoPowerSupplyRedundancyandcheckedEnableDynamicPowerSupplyEngagement.
FortheHPBladeSystemc-Class,wetestedwith"PowerSavings"enabledandthepowersuppliessetto"ACRedundant.
"FortheIBMBladeCenterHType8852,ourexperimentshadnosignificanteffectonthepowersavings.
Beforedeterminingthefinalsettingsforeachenclosure,weranasingletestwitheachenclosure'smaximumnumberofblades;thePowersettingssectionshowsthoseresults.
TheeffectontheresultsoftheDellandIBMwaslessthan2%inallcases.
Wetestedwiththefollowingdefaultsettingsforeachenclosure:DellPowerEdgeM600:ServerPowerThrottlingEnabled:CheckedRedundancyPolicy:PowersupplyredundancyEnableDynamicPowerSupplyEngagement:Checked11PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversHPBladeSystemc-Class:EnclosurePowerMode:ACredundantEnableDynamicPowerSavingsMode:CheckedIBMBladeCenterHType8852:Domain1:Non-redundantDomain2:Non-redundantAcousticmode:DisabledPowermeasurementprocedureTorecordeachbladesystem'spowerconsumptionduringtesting,weusedanExtechInstruments(www.
extech.
com)380803PowerAnalyzer/Datalogger.
Weconnectedthepowercordfromthesystem'spowersupplytothePowerAnalyzer'soutputloadpoweroutlet.
WethenpluggedthepowercordfromthePowerAnalyzer'sinputvoltageconnectionintoapoweroutlet.
Weusedthissetupforeachpowersupplyinthechassis.
Becauseeachofthethreeservershas6powersupplies,weused6ExtechPowerAnalyzersfortesting.
WeconnectedallExtechPowerAnalyzerstoonemonitoringsystemtorecordthepowerdrawofthesystems.
WeusedthePowerAnalyzer'sDataAcquisitionSoftware(version2.
11)tocaptureallrecordings.
WeinstalledthesoftwareonaseparatePC,towhichweconnectedallPowerAnalyzersviaaseparatedRS-232cableforeachExtech.
Wecapturedpowerconsumptionat1-secondintervals.
Togaugetheidlepowerusage,werecordedthepowerusagefor2minuteswhileeachserverwasrunningtheoperatingsystembutotherwiseidle.
Tocomputethetotalpower,wetookthewattagesumfromeachofthemeters.
Weaveragedthepowerusageduringtheperiodtheserverwasrunningthebenchmark.
Wecallthistimethepowermeasurementinterval.
SeeFigures2(idlepowerconsumption)and3(averagepeakpower)fortheresultsofthesemeasurements.
SPECjbb2005configurationWeusedSPECjbb2005version1.
07,datedMarch15,2006.
WefollowedSPEC'srunrules.
(FormoreinformationaboutSPECjbb2005anditsrunrules,seewww.
spec.
org/jbb2005/docs/RunRules.
html.
)WeinstalledSPECjbb2005bycopyingthecontentsoftheSPECjbb2005CDtothedirectoryC:\SPECjbb2005v1.
07ontheserver'sharddisk.
SPECjbb2005requiresaJavaVirtualMachineonthesystemundertest.
WeusedtheBEAJRockit(R)(buildP27.
4.
0-10-90053-1.
6.
0_02-20071009-1827-windows-x86_64,compiledmode)JVMforthistestingandleftthedefaultinstallationsettings.
Afterinstallation,aspertherunrules,weeditedtheSPECjbb_config.
propsfileintherootSPECjbb2005directorytoincludedisclosureinformationabouttheserverandourlicenseinformation.
SPECjbb2005usesthisfilewhengeneratingtheresultsoutputforeachrun.
WealsomodifiedtheSPECjbb.
propsfiletochangethenumberofJVMinstancesto2.
Thischangeallowsaservertorun2JVMinstancesduringtesting.
Wecreatedabatchfile,whichweplacedintherootSPECjbb2005directory,toissuetheJavaruncommandtolaunchthebenchmark.
Duringtesting,weusedthecommandpromptwindowwithinMicrosoftWindowsServer2003x64Editiontorunthisbatchfile,thetextofwhichisasfollows:12PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeservers@echooffsetpath="C:\jrockit-jdk1.
6.
0_02\bin";%path%setJVM=2::SetJAVA_HOMEtoJava.
exepath.
setJAVA_HOME="C:\jrockit-jdk1.
6.
0_02\bin":stage1setPROPFILE=SPECjbb.
propssetJAVAOPTIONS=-Xms256m-Xmx256mremsetJBBJARS=.
\jbb.
jar;.
\check.
jarsetJBBJARS=.
\jbb.
jar;.
\jbb_no_precompile.
jar;.
\check.
jar;.
\reporter.
jarsetCLASSPATH=%JBBJARS%;%CLASSPATH%:stage2echoUsingCLASSPATHentries:for%%cin(%CLASSPATH%)doecho%%c@echoonstart/bC:\jrockit-jdk1.
6.
0_02\bin\java.
exe%JAVAOPTIONS%spec.
jbb.
Controller-propfile%PROPFILE%@echooffsetI=0setJ=F:LOOPset/aI=%I+1echo.
echoStartingJVMNumber%I%withAffinitytoCPU%J%echo.
@echoonstart/AFFINITY%J%/BC:\jrockit-jdk1.
6.
0_02\bin\java.
exe-Xms1600m-Xns1300m-Xmx1600m-XXaggressive-XXlargepages-XXcallprofiling-Xgc:genpar-XXthroughputCompaction-XXlazyUnlocking-XXtlasize:min=4k,preferred=256kspec.
jbb.
JBBmain-propfile%PROPFILE%-id%I%>multi.
%I%@echooffsetJ=%J%0IF%I%==%JVM%GOTOENDGOTOLOOP:END:egress13PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversInthebatchfilewesettheJavaoptionsthatcontroltheperformanceoftheJVMasfollows:Xms1600mThisoptionsetstheminimumheapsize.
Wesettheminimumandmaximumheapsizestobethesame,sotheheapsizewouldstayaconstant1600MB.
Xns1300mThisoptionsetsthenurserysizeto1300MB.
Xmx1600mThisoptionsetsthemaximumheapsize.
XXaggressiveThisoptionessentiallytellstheJVMtoperformatmaximumspeed.
Xgc:genparThisoptiontellsJavatousegenerationalparallelgarbagecollection.
XXthroughputCompactionThisoptionadjuststhecompactionratiodynamicallybasedonlivedataintheheap.
XXlazyUnlockingThisoptiondetermineswhentheJVMreleaseslocks.
XXtlasize:min=4k,preferred=256kThisoptionsetsthethread-localareasizetheJVMuses.
Wespecifiedaminimumandpreferredsettingfortesting.
-XXlargepagesThisoptiontellstheJVMtouselargepages,iftheyareavailable,fortheJavaheapandotherareasintheJVM.
-XXcallprofilingThisoptionenablestheuseofcallprofilingforcodeoptimizations.
TestexecutionForustocalculatetheaveragepowerduringpeakperformance,weneededallsystemstoberunningatmaximumperformanceatthesametime.
Toachievethis,weneededallbladeserverstostartSPECjbb2005atthesametime,whichweaccomplishedbyusingbatchfilestostartSPECjbb2005onallsystemsundertest.
Oneachsystemundertest,wecreatedabatchfileinthestartupfolderthatwouldstartassoonastheoperatingsystemloadedandthensleep,orsitidle,for720seconds.
After720seconds,thebatchfilewouldsearch,oncepersecond,forarun.
txtfileintheSPECjbb2005directory.
Tobeginthetest,weusedabatchfileonacontrollersystemthatcopiedtherun.
txtfiletoallsystems.
Oncethatbatchfilehadcopiedtherun.
txtfiletothesystemsundertest,therunningbatchfileswouldstarttheSPECjbb2005benchmark.
BystartingSPECjbb2005thisway,weensuredthatallclientsstartedwithin1secondofeachother.
14PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversAppendixA–EnclosureconfigurationinformationFigure17providesdetailedconfigurationinformationabouttheenclosures,whichwepresentinalphabeticalorder.
EnclosureDellPowerEdgeM600HPBladeSystemc-ClassIBMBladeCenterHType8852GeneraldimensioninformationHeight(inches)17.
517.
515.
75Width(inches)191919Depth(inches)30.
531.
028.
0Usizeinserverrack10109Numberofblades161614PowersuppliesTotalnumber662Wattageofeach2,3602,2502,900CoolingfansTotalnumber9102blowersDimensions(HxW)ofeach3.
5x33.
5x34.
5x11.
5Voltage12volts12volts200-240voltsAmps7amps16.
5amps5.
5ampsFigure17:Detailedconfigurationinformationabouttheenclosures.
15PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversAppendixB–BladesystemconfigurationinformationFigure18providesdetailedconfigurationinformationaboutthebladeserversystems,whichwepresentinalphabeticalorder.
ServersDellPowerEdgeM600HPBladeSystemc-ClassIBMBladeCenterHType8852GeneralprocessorsetupNumberofprocessorpackages222Numberofcoresperprocessorpackage444Numberofhardwarethreadspercore111SystemPowerManagementPolicyAlwaysonAlwaysonAlwaysonCPUVendorIntelIntelIntelNameQuad-CoreIntelXeonprocessorE5345Quad-CoreIntelXeonprocessorE5345Quad-CoreIntelXeonprocessorE5345SteppingB77SockettypeLGA771LGA771LGA771Corefrequency(GHz)2.
33GHz2.
33GHz2.
33GHzFront-sidebusfrequency(MHz)1,333MHz1,333MHz1,333MHzL1cache32KB+32KB(percore)32KB+32KB(percore)32KB+32KB(percore)L2cache2x4MB(each4MBsharedbytwocores)2x4MB(each4MBsharedbytwocores)2x4MB(each4MBsharedbytwocores)PlatformVendorandmodelnumberDellPowerEdgeM600HPProLiantBL460cBladeCenterHS21MotherboardmodelnumberDell0MY736HP435458-B21IBM8853C2UMotherboardchipsetIntel5000PIntel5000PIntel5000PMotherboardrevisionnumberX3191B1BIOSnameandversionDell0.
2.
11HPI1512/26/2006IBM1.
0710/26/2007BIOSsettingsDisabledHardwarePrefetcherandAdjacentCacheLinePrefetcherDisabledHardwarePrefetcherandAdjacentCacheLinePrefetcherDisabledHardwarePrefetcherandAdjacentCacheLinePrefetcherChipsetINFdriverIntel7.
4.
1005HP2.
1.
8Intel7.
4.
1005Memorymodule(s)VendorandmodelnumberSamsungM395T2953EZ4-CE65MicronMT18HTF12872FDYHynixHYMP512F72CP8D2-Y5TypePC2-5300PC2-5300PC2-5300Speed(MHz)667MHz667MHz667MHzSpeedinthesystemcurrentlyrunning@(MHz)667MHz667MHz667MHzTiming/Latency(tCL-tRCD-iRP-tRASmin)5-5-5-155-5-5-155-5-5-15Size4GB(4x1GB)4GB(4x1GB)4GB(4x1GB)NumberofRAMmodules444ChiporganizationDualsideDualsideDualside16PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversServersDellPowerEdgeM600HPBladeSystemc-ClassIBMBladeCenterHType8852HarddiskVendorandmodelnumberFujitsumay2073rcSeagateSt973402SSIBM26K5777Numberofdisksinsystem222Size73GB72GB73.
4GBBuffersize16MB16MB8MBRPM10,00010,00010,000TypeSASSASSASControllerDellSAS6/iRIntegratedBladesControllerSmartArrayE200IcontrollerLSIAdapter,SAS3000seriesControllerdriverDell1.
24.
4.
0HP6.
6.
0.
64LSI1.
21.
28.
0OperatingsystemNameMicrosoftWindowsServer2003,Enterprisex64EditionMicrosoftWindowsServer2003,Enterprisex64EditionMicrosoftWindowsServer2003,Enterprisex64EditionBuildnumber379037903790ServicePackSP2SP2SP2MicrosoftWindowsupdatedateSP2plusupdatesthrough11/29/07SP2plusupdatesthrough11/29/07SP2plusupdatesthrough11/29/07FilesystemNTFSNTFSNTFSKernelACPIMultiprocessorx64-basedPCACPIMultiprocessorx64-basedPCACPIMultiprocessorx64-basedPCLanguageEnglishEnglishEnglishMicrosoftDirectXversion9.
0c9.
0c9.
0cGraphicsVendorandmodelnumberATIES1000ATIES1000ATIES1000ChipsetATIES1000ATIES1000ATIES1000BIOSversionBK-ATIVER008.
005.
031.
000BK-ATIVER008.
005.
013.
000BK-ATIVER008.
005.
031.
000TypeIntegratedIntegratedIntegratedMemorysize32MB32MB16MBResolution1,024x7681,024x7681,024x768DriverATI8.
240.
50.
1000ATI8.
24.
3.
0ATI8.
24.
3.
0Networkcard/subsystemVendorandmodelnumberBroadcomBCM5708SNetXtremeIIGigEHPNC373iMultifunctionGigabitServerAdapterBroadcomBCM5708SNetXtremeIIGigETypeIntegratedIntegratedIntegratedDriverBroadcom3.
5.
8.
0HP3.
0.
5.
0Broadcom3.
0.
5.
0OpticaldriveVendorandmodelnumberNoneinstalledNoneinstalledNoneinstalledUSBportsNumber22(withadapterattached)2TypeUSB2.
0USB2.
0USB2.
0Figure18:Detailedconfigurationinformationaboutthebladeserversystems.
17PrincipledTechnologies,Inc.
:SPECjbb2005performanceandpowerconsumptiononDell,HP,andIBMbladeserversDisclaimerofWarranties;LimitationofLiability:PRINCIPLEDTECHNOLOGIES,INC.
HASMADEREASONABLEEFFORTSTOENSURETHEACCURACYANDVALIDITYOFITSTESTING,HOWEVER,PRINCIPLEDTECHNOLOGIES,INC.
SPECIFICALLYDISCLAIMSANYWARRANTY,EXPRESSEDORIMPLIED,RELATINGTOTHETESTRESULTSANDANALYSIS,THEIRACCURACY,COMPLETENESSORQUALITY,INCLUDINGANYIMPLIEDWARRANTYOFFITNESSFORANYPARTICULARPURPOSE.
ALLPERSONSORENTITIESRELYINGONTHERESULTSOFANYTESTINGDOSOATTHEIROWNRISK,ANDAGREETHATPRINCIPLEDTECHNOLOGIES,INC.
,ITSEMPLOYEESANDITSSUBCONTRACTORSSHALLHAVENOLIABILITYWHATSOEVERFROMANYCLAIMOFLOSSORDAMAGEONACCOUNTOFANYALLEGEDERRORORDEFECTINANYTESTINGPROCEDUREORRESULT.
INNOEVENTSHALLPRINCIPLEDTECHNOLOGIES,INC.
BELIABLEFORINDIRECT,SPECIAL,INCIDENTAL,ORCONSEQUENTIALDAMAGESINCONNECTIONWITHITSTESTING,EVENIFADVISEDOFTHEPOSSIBILITYOFSUCHDAMAGES.
INNOEVENTSHALLPRINCIPLEDTECHNOLOGIES,INC.
'SLIABILITY,INCLUDINGFORDIRECTDAMAGES,EXCEEDTHEAMOUNTSPAIDINCONNECTIONWITHPRINCIPLEDTECHNOLOGIES,INC.
'STESTING.
CUSTOMER'SSOLEANDEXCLUSIVEREMEDIESAREASSETFORTHHEREIN.
PrincipledTechnologies,Inc.
1007SlaterRoad,Suite250Durham,NC27703www.
principledtechnologies.
cominfo@principledtechnologies.
comPrincipledTechnologiesisaregisteredtrademarkofPrincipledTechnologies,Inc.
Allotherproductnamesarethetrademarksoftheirrespectiveowners