differingisaserver

DeployingF5NetworksBIG-IPLocalTrafficManagerwithWebAcceleratorforMicrosoftOfficeSharePointServer2007Introduction.
2Audience.
3Solutionoverview3Technologies4Branchofficescenario5Internetscenario6Examplesolutionconfigurations.
6Softwarecomponents6HPservers.
6WANaccelerationhardware7Solutiondeployment.
8Installation.
8Solutionset-up8Examplesetupscreens.
9Asymmetricoperation.
12Symmetricoperation.
13Performanceresults14Testmethodology.
14Workloadscenarios14Networkconfigurationsandparameters15Datatablemetrics16Branchofficeresults17Internetresults.
18Keyfindings19Summary20HPServicesandF5solutions.
20Implementingaproof-of-concept.
21Formoreinformation.
22IntroductionOrganizationsareusingMicrosoftOfficeSharePointServer2007todeployintelligentportalsthatprovideconnectivitybetweenpeople,processes,customers,andintellectualcapital.
Specificexamplesincludebusinessprocesssolutionssuchasteamcollaboration,datarecordsmanagement,documentmanagementandpublishing,anddynamicscore-carding.
Thegrowthrateisincreasing,withadoptionofOfficeSharePointServer2007whichprovidesincreasedfunctionalityatsomewhathigherresourcecosts(serverCPU,memory,network,etc.
)thanSharePoint2003.
ManyorganizationswhoaremovingtothenewversionarealsotakingtheopportunitytoconsolidatetheirdistributeddeploymentsandSharePointfarmsintofewercentralizeddatacenters.
Thisapproachhasmanyprovencostbenefits,butrequiresthesenewdatacentersbeabletoprovideconsistent,predictable,andadequateservicetouserswhoarenowlocatedmoreremotelyfromtheSharePointfarmsacrossacorporateWideAreaNetwork(WAN).
Indistributedfarmdeployments,servercapacity(usuallytheWebfront-end(WFE)servers)waslikelythelimitingresourcetoavailablefarmcapacity,withlocalusersbeingconnectedoverthecorporateLocalAreaNetwork(LAN)atbackbonebandwidthsof100Mbit/secoreven1Gbit/sec.
Bymovingthefarmstoacentralizedlocation,potentiallimitstocapacityandperformancebasedonavailableWANbandwidth(forexample,aT3connectionat44Mbit/sec,orevenaT1connectionat1536Kbit/sec)canbecreated.
Further,theintroductionoflatency(delay)duemainlytoconnectiondistance,andsomesmallpercentageofdroppedpacketswillhaveanimpactonperformance.
WhyhasthenetworknowbecomemoreimportantFigure1showsexamplenetworktrafficbetweenendusers,aSharePointWFEserver,andSharePointdatabaseserver(MicrosoftSQLServer).
ThenetworktrafficdataobservedintheHPSolutionAlliancesEngineering(SAE)NashuaLabswasrunningatathroughputrateofabout20requestspersecond(RPS).
Eachsuchrequestrepresentsauser'clicking'onaportalpagetoperformafunction(examples:homepage,search,check-indocument,downloaddocument,openadocumentlibrary,list,etc.
).
ThusdividingtheKB/secvaluesinFigure1by20generalizestheaveragenetworkresourcecost(trafficrate)ofasingleuserfunction.
Thevalueswillvarybasedonexactworkloadfunctionsandmixes,documentsizes,etc.
,butarearepresentativeexample.
Figure1.
ExampleNetworkTrafficratesat20RPSthroughputThedual-coreHPc-ClassBladeSystemWFEserverstested(seesectionExamplesolutionconfigurations–HPservers)caneachachieveabout80RPS(runningWFEandQuerySearchservices),andtestedquad-coreserversabout160RPSperWFE.
ThusthenetworktrafficbetweentheWFEandusersatasingleservercapacityof80RPScanbemorethan25000KB/sec,orabout40Mbps.
Thistraffic,producedbyasinglepowerfulWFEserver,thusexceedstheusablecapacityofaT3WAN(44Mbps).
AstheWANaveragebandwidthusageexceedsabout75%andapproachessaturation,userresponsetimeswillincreasedramatically,throughputwillfall,userstypicallywillbecomefrustrated,andthesolutionbecomesinefficient.
Technologiestoreduce(optimize)WANbandwidthuseandminimizetheeffectsoflatencyarethushighlydesirable.
OneapproachtoimprovingoverallsystemperformanceistheuseofaclassofTrafficManagementDevicecommonlyreferredtoasaWANAccelerator.
Thispaperdetailstheresultsofusingonesuchsolution–theF5NetworksBIG-IPLocalTrafficManager(LTM).
ByenablingtheWebAcceleratormoduleonBIG-IPLTM,testresultsshowedasignificantincreaseinuser-perceivedperformancewhileatthesametimedecreasingtheloadontheWFEs.
ThistechnologywastestedintheHPSAENashuaLabsintwotypicalcustomerscenarios.
A"branchoffice"scenariorepresentingcorporateusersaccessingSharePointperformingread/writeoperations;andan"Internet"scenariorepresentinganonymoususersaccessing(readonly)acompany'sexternally-facingInternetportal.
Thetwosolutiondeployments,best-practiceconfigurationsandobservedperformanceresultsaredetailedinthefollowingsections;anddemonstratehowsignificantperformanceimprovementscanberealized.
NOTE:Throughoutthisdocument,allreferencestoBIG-IPrefertoBIG-IPLocalTrafficManager(LTM)withtheWebAcceleratormoduleenabled.
AudienceThispaperisintendedforpeoplewhowillbeproposingsolutions,providinginstallationservicesorconsulting,andwhomaybeassistingindeployingOfficeSharePointServer2007solutionsaccessedoveraWideAreaNetwork(WAN).
ItwillalsobeofinteresttoITprofessionalswhomaybedeployingand/ormanagingsuchOfficeSharePointServer2007solutions.
HPrecommendstheinformationprovidedhereinbeusedinconjunctionwiththeOfficeSharePointServer2007productsolutiondocumentation;andinformationcontainedinadditionalwhitepapersandarticlesauthoredbytheWANtechnologysolutionvendor,F5NetworksInc.
,asnotedinthesectiontitledFormoreinformation.
SolutionoverviewThesolutionsdescribedemployF5BIG-IPLTMwithWebAcceleratorappliancesinbothsymmetricandasymmetricdeployments,asapplicableforthetwoexamplescenarios(branchofficeandInternet).
Forthistestengagement,HPSAELabsselectedtouseBIG-IPLTMwiththeWebAcceleratorsoftwaremodule.
BIG-IPLTMprovidestheloadbalancingfunctionalitytoensurehighavailabilityandscalability,whileWebAcceleratorfocusesonincreasingoverallHTTPandHTTPSperformance.
Althoughthetwoproductscanbepurchasedasseparatedevices,thecombinedfunctionalityprovidesabest-of-breedsolutionatacosteffectiveprice.
TheF5BIG-IPLTMwithWebAcceleratorplatformselectedwasthe6800,whichisa2U,portdensenetworkappliancecapableofhandlingnearly4Gbpsofsustainedthroughput.
3ThehardwarerunsF5'sTMOS(TrafficManagementOperatingSystem)architectureplatform,whichwasbuiltfromthegroundupasanOSwhosesolepurposewasapplicationdeliverynetworking.
Theendresultisaplatformthatdeliversunparalleleddatainspection,manipulation,andaccelerationatlinespeed.
Withmorethan16,000customers,F5hasmorethan60%oftheworldwidemarketshareoftheadvancedplatformApplicationDeliveryControllermarketfor2Q07basedonrevenue(source:Gartner,MarketShare:ApplicationAccelerationEquipment,Worldwide,2Q07,October2007).
Inasymmetricdeployment,twosuchBIG-IPLTMwithWebAcceleratorapplianceslocatedateach'edge'oftheWANcanworkcollaborativelytoprovidesignificantWANtrafficreductionandoptimization.
Asingleapplianceinanasymmetricdeploymentcannotprovideallofthefeaturesofasymmetricconfiguration,butcanstillresultinnotableimprovementsinbandwidthusageanduser-perceivedperformance.
TechnologiesTheBIG-IPLTMwithWebAcceleratorapplianceprovidesanumberoftechnologiesandfeaturesthatcanbeenabledaspartofaSharePointdeployment.
Thesetechnologiesandfeaturesaredesignedtoconsistentlydeliverydatatotheenduserinthefastestmethodpossiblewhilealsodecreasingserverandbandwidthloads.
Theyinclude:HardwareassistednetworkroutingandbridgingTrafficmanagement(forexample,WFEserverloadbalancing)–Applicationhealthmonitoring–ServerperformancebasedloadbalancingalgorithmsHTTP/Sapplicationaccelerationtechnologies–Dynamicendpointcaching–Intelligentcompression–SSLacceleration–TCPmultiplexing–IntelligentBrowserReferencing(IBR)–SharePointlayer7(object)protocoloptimizationTCPoptimizations(optimizesconnectionsforbothlocalandremoteusers)–Delayedandselectiveacknowledgements–Explicitcongestionnotification–TCPslowstart–Congestionavoidance–TCPSpoolingThesecapabilitiescanbeselectedasappropriateineachscenariotomaximizetheefficiencyoftheWAN.
ThebestpracticeforconfiguringtheseoptionsistoutilizetheBIG-IPWebAccelerator"SharePointprofile"(seelatersectionSolutionset-up).
Asnotedearlier,itisthetrafficfromtheWFEstotheusers'desktopbrowsersthatcomprisesthevastmajorityoftheWANtraffic.
ThistrafficcanincludeHTTP,Java,XML,etc.
toprovidepagecontent,andvariousMicrosoftOfficedocumentsorotherfiles'opened'bytheuser.
Reductioninthattrafficbyoptimization,compression,andbyclient-endcachingwillimproveperformanceindifferingdegreesinbothexamplescenarios.
4WhentailoredtoaspecificSharePointdeployment,theprecedingtechnologiesshouldprovide:HighAvailability–BIG-IPLTMwithWebAcceleratorisacriticalpieceofprovidingahighlyavailableSharePointinfrastructure.
ByconfiguringBIG-IPLTMwithWebAcceleratortomonitortheWFEsinordertodetermineiftheyareservingvalidcontentornot,userswillonlybesenttotheserversthataretrulyavailable.
AproperlyconfiguredBIG-IPLTMwithWebAcceleratorwillnotsendanyuserstoaWFEthatismis-configured,down,orsaturated.
ByimplementingtheoverallsolutionwithredundancyintheBIG-IPLTMwithWebAcceleratorappliances,SharePointServersandServices,aswellasaSQLclusteredback-end,itispossibletobuildasystemthatcantrulywithstandthelossofanyonecomponent.
Scalability–Asenterprisesgrow,itiscriticalthatinformationsystemscanscaletoservetheneedsofitsusers.
Singleserverdeploymentscanscalebybuildinghigher-endsystems;howeverthereisameasurablepointinwhichthisbecomesacostineffectivemodel.
BydeployingBIG-IPLTMwithWebAccelerator,itispossibletoscaleoutmultipleWFEsandspreadtheuserloadamongstthefarm.
ByoptimizingthetrafficviatechnologieslikecachingandTCPmultiplexing,youcanactuallyscalethesystemwithoutexperiencingserverproliferationorbandwidthsaturation.
Performance–Countlessstudieshaveshownthatuseradoptionofacorporateresourcewillsufferasuserperceivedperformanceofthesystemdeclines.
Inordertomakesurethatallusers,whetherlocalorremote,havethebestoveralluserexperience,itwillbehighlybeneficialtoleverageBIG-IPLTMwithWebAcceleratorSharePointAccelerationfunctionality.
BranchofficescenarioThebranchofficescenariorepresentsacentralizeddatacentersupportingOfficeSharePointServer2007farms,andanumberofauthenticatedusersco-locatedataremotelocationconnectedbyacorporateWANofknownbandwidth,latency,etc.
TheWANcharacteristicswilllikelydependongeographiclocationandspan;andthenetworkusagewilldependonthenumberofusersattheremotelocationandtheworkload(functionsandfrequency)thattheyapplytosupporttheirbusinessneeds(SharePointandothertraffic).
ABIG-IPLTMwithWebAcceleratorsymmetricdeploymentbestsuitsthisscenario.
BIG-IPLTMwithWebAcceleratorapplianceswillbelocatedbothatthedatacenterandattheremotelocation,andbeconfiguredtoworkco-operatively.
Thisconfigurationcanprovideallthefeaturesnotedabove,includingveryhighratiocompression/decompressionofcontentbytheappliances.
Theremoteappliancecanalsointelligentlycachecontentlocaltotheusers,thusgreatlyreducingWANtraffic.
Anappliancecanbelocatedateachsuchremoteofficelocationasneeded.
Thenumberofappliancesatthedatacenterwilldependontheappliancecapacities(differentmodelsprovidingvaryingcapacity)andonhigh-availability(fail-over)needs.
5InternetscenarioTheInternetscenariodepictsacentralizeddatacenterprovidingOfficeSharePointServer2007sitesbeingaccessedbyanonymoususersontheInternet(outsidethecorporatenetworkperimeter).
Theseuserswillbeconsumersofinformation(read-onlyaccess)andthecorporatenetworkwillbeprotectedbyappropriatesecurity(forexample,deploymentofMicrosoftInternetSecurityandAcceleration(ISA)Server)atthenetworkperimeter(edgeoftheWAN).
Asymmetricconfigurationisnotfeasibleinthiscase,thusanasymmetricsingleBIG-IPLTMwithWebAcceleratorappliancewillbeusedatthedatacenter,locatedbehindthenetworkperimeter.
ThisconfigurationcannotachieveallthecapabilitiesofasymmetricBIG-IPLTMwithWebAcceleratordeployment,howeveritstillprovidesagooddegreeofcompressionefficiencyasitcanusetheusers'browserstodecompresscontent;andcachingefficiencyisstillhighastheBIG-IPWebAcceleratorIntelligentBrowserReferencing(IBR)technologywillensuretheusers'browsercachesaremaintainedcurrent.
TestresultsshowanasymmetricdeploymentprovidesnotableWANoptimizationandimprovedperformance.
ExamplesolutionconfigurationsThefollowingsectionsprovidedetailsofthesolutioncomponentsandconfigurationsastestedintheHPSAELabs.
SoftwarecomponentsThefollowingsoftwarecomponentswereutilizedonthevariousOfficeSharePointServersinthefarm,eachaseitherX86(32-bit)orX64(64-bit)versionsasneeded:WindowsServer2003SP1(onallservers)MicrosoftSQLServer2005SP2(ondatabaseservers)MicrosoftOfficeSharePointServer2007Onceallsoftwarewasinstalled,theSharePoint2007ConfigurationWizardwasusedtocreatethefarmAdministrationservice,andthentherequiredWFE,QuerySearch,andIndexSearchservices.
Testfarmsiteswerecreatedandcontentdeployedtodocumentlibraries,listscreated(forexample,events,announcements),andothersitepagesandcontenttosupportthestandardHPtestworkloads(describedlater).
HPserversThetestfarmcomprisedatotalofsevenHPc-Classbladeserversrepresentinganenterprise-classsolutiondeployment,configuredas:2xHPProLiantBL465c,Dual-CoreAMDOpteron,2.
80GHz–WFE+QuerySearch2xHPProLiantBL460c,Quad-CoreIntelXeon,3.
0GHz–WFE+QuerySearch1xHPProLiantBL460c,Quad-CoreXeon,3.
0GHz–IndexSearch2xHPProLiantBL460c,Quad-CoreXeon,3.
0GHz–Databaseservers(active/passiveSQLcluster)6Asnoted,thefirstfourserverswererunningboththeWFEandQuerySearchservicesandwereconfiguredasaload-balancedpooltotesttheBIG-IPLTMwithWebAcceleratorloadbalancingcapability(asopposedtousingWindowsLoadBalancingService).
Normallyserversofthesame'power'wouldbeused,howeverBIG-IPLTMwithWebAcceleratorprovidesmultipleloadbalancingmethodsthatcaneffectivelybalanceloadacrossserversofdisparatepower(seethesectiontitledSolutionsetupforconfigurationdetails).
AsthethroughputofasingleWFEofthispowercanexceedtheusablenetworkcapacityofa44MbpsWAN,theWANtestsranagainstasingleBL460cquad-coreserver.
Thisalsoprovidedasingleconsistenttestplatformforalltests.
WANaccelerationhardwareFigure2depictstwoBIG-IPLTM6800withWebAcceleratorappliancesdeployedinasymmetricconfiguration.
AuthenticatedUserswereconnected(viaanetworkswitch)totheclient-sideBIG-IPonaremoteLAN.
Theserver-sideBIG-IPLTMwithWebAcceleratorwasconnectedtotheHPc-Classenclosureinternalnetworkswitch.
TheotherportsoftheBIG-IPLTMwithWebAcceleratorapplianceswereconnectedtotheWAN(aWANemulatorwasemployedintheHPSAElabtests).
Figure2.
SymmetricF5BIG-IPLTMwithWebAcceleratordeploymentFigure3showsasingleBIG-IPLTM6800withWebAcceleratorapplianceconnectedtotheHPc-Classbladeenclosureinternalnetworkswitch.
TheotherBIG-IPportwasconnectedtoanISAserver(providingcorporateLANsecurity).
AnonymoususersconnectedviaanetworkswitchtotheWAN,andthentotheISAserver.
7Figure3.
AsymmetricF5BIG-IPLTMwithWebAcceleratordeploymentThefollowingdescribesbest-practicesolutiondeploymentguidelines,andBIG-IPLTMwithWebAcceleratorconfigurationdetailsastestedintheHPSAElabs.
SolutiondeploymentThesesectionssummarizekeyaspectsoftypicalsolutiondeployments,althoughindividualcustomerdeploymentspecificsmayvary.
Fordetailedinformation,refertotheF5BIG-IPSharePoint2007ApplicationReadyNetworkDeploymentGuide,apointertowhichisprovidedinthesectionFormoreinformation.
InstallationPhysicalinstallationoftheBIG-IPLTMwithWebAcceleratorappliancesissimple,eachrequiringa2Urackspaceandpowertotherear-mountedpowersupplysocket(s).
Inmostcases,theBIG-IPLTMwithWebAcceleratorapplianceswillbeimplementedinaredundantfashion,withtwounitsoperatinginanactive/standbypair.
An'outofband'managementnetworkportonthefrontoftheBIG-IPLTMwithWebAcceleratorprovidessecureaccesstotheBIG-IPsubsystem.
Thisportisconnectedtoadedicatedonboardprocessorthatcanprovidelights-outmanagement,monitoring,anddiagnosticsoftheBIG-IPLTMwithWebAcceleratorappliance.
Basicmanagementandmonitoringcanalsobeachievedviathedisplayscreenonthefrontoftheunit.
Networkcablesforproductiontrafficcanbepluggedintothetri-speedcopper/fibreportsfoundonthefrontoftheBIG-IPLTMwithWebAcceleratorappliance.
Theseportsareconnectedviaatruelayer-2switchplane,andallowforthetraffictopassthroughthetrafficmanagementengineasnecessary.
Solutionset-upTheBIG-IPLTMwithWebAcceleratormanagementportshipswithadefaultIPaddress,howeveritiseasilymodifiedviathefrontpanelinordertomatchyournetwork.
TheBIG-IPadministratorcanusethatIPaddresstoaccessthecomprehensivebrowser-basedconfigurationpageviaHTTPSandapowerfulcommand-lineinterface(CLI)viaSSH.
Furthersetupandconfigurationisperformedviavarioushierarchicalscreensinthewebinterface.
8F5hasdocumentedthestep-by-stepproceduresforconfiguringtheBIG-IPLTMwithWebAcceleratorsystemtomanageandacceleratetrafficinaSharePoint2007environmentintheirF5BIG-IPSharePoint2007ApplicationReadyNetworkDeploymentGuide,apointertowhichisincludedinthesectionFormoreinformation.
Pleaserefertothisdocumentfordetailedsetupprocedures.
ThefollowingwilldescribeanexampleSharePoint2007andBIG-IPLTMwithWebAcceleratordeployment(astestedintheHPSAELabs)byshowingsomekeyBIG-IPadministrationscreensandexplainingtheirpurpose.
Althoughnotanexhaustiveconfigurationguide,itwillshowsomeofthesettingsthatshouldbeconfiguredinatypicaldeployment.
ThespecificBIG-IPLTMwithWebAcceleratorfeaturesemployedasbest-practicesinasymmetricandsymmetricdeploymentsarealsodiscussed.
ExamplesetupscreensFigure4showstheVirtualServersPropertiestab.
TheGeneralPropertiesshowavirtualservername(VS_Wan),adestinationhostIPaddressandtheport(HTTPport80)beingmonitored.
TheConfigurationsectionshowsthattheunderlyingprotocolisTCP,andthattheHTTPProfile"http-wan-optimized-compression"isbeingused.
Figure4.
BIG-IPLTMwithWebAcceleratorsetup–VirtualServergeneralproperties9Figure5showstheVirtualServerProfilePropertiespage.
ThisenablesdefinitionofaClassbasedonanumberofprovidedprofileswhichcanbecustomized.
OfnoteistheselectionintheConfigurationsectionshowingWebAcceleratorisenabled(asnotedabove).
Theselectionof"WebAcceleratorEnabled"intheprofilepropertiesprovidesclient-endcachingfunctionality,thususingtheHTTPprofile"http-wan-optimized-compression"perFigure4(asopposedto"http-wan-optimized-compression-cache")isthecorrectselection.
Figure5.
BIG-IPLTMwithWebAcceleratorsetup–VirtualServerprofileproperties10Figure6showstheVirtualServerresourcesscreen.
Thisenablesselectionoftheloadbalancingpool("SharePointFrontEndspool"showndefinedinFigure7)andoftheHTTPClassProfile("WAN_Class",asshowndefinedinFigure5above).
Figure6.
BIG-IPLTMwithWebAcceleratorsetup–VirtualServerresources11Figure7showstheLoadBalancingpooldefinitionasregardsthepoolserverIPaddressesandtheloadbalancingmethodchoices.
Variousserverconfigurationsandloadbalancingmethodsweresuccessfullytested.
Balancingserversofdisparatepowerispossiblebyapplyingweightingtotheserverstoreflecttheirindividualpowerorthroughputcapacity.
Intheexamplebelow,onlyoneserver(IPaddress129.
3.
30.
113:80)isshownwithagreen'activeandenabled'symbol,assubsequentWANtestingwasperformedusingonlythatoneserverasnotedearlier.
Theblackdiamondsymboldenotesan'inactive'serverthatcouldnotbecontacted(poweredoff),andtheblackcirclesymbolsrepresentactiveservers(canbecontacted)butarenotcurrentlymembersoftheloadbalancedpool(disabled).
Figure7.
BIG-IPLTMwithWebAcceleratorsetup–VirtualServerpoolmembersThefollowingsectionsprovidedetailsregardingtheBIG-IPLTMwithWebAcceleratorfeaturesprovidedwithasymmetricdeployment;andadditionalfeaturesfromsymmetricdeployment.
AsymmetricoperationByfront-endingtheSharePointWFEswithBIG-IPLTMwithWebAccelerator,remoteusersconnectingtothedatacenterforcontentwillnoticeasignificantdecreaseinpageretrievaltime.
Thereareseveraltechnologiesthatmakethishappen.
Theseinclude(butarenotlimitedto):BIG-IPWebAcceleratorDynamicCaching–Cachesunchangingdatathatmayseemdynamic(containsqueryparameters,etags,sessionIDs),butisactuallystaticdataorchangesinanidentifiablepattern.
BIG-IPWebAcceleratorcancacheahigherpercentageofdatafrom12dynamicwebapplicationswhilemaintainingproperapplicationbehavior.
ItaccomplishesthisbyfullyinspectingeveryaspectofHTTPrequests,controllingcachingbehavior,andinvalidatingcacheddata.
BIG-IPLTMSSLAcceleration–Offloadstheserversfromcomputational-intensiveSSLencryptionanddecryption,reducingserverprocessorutilizationbyasmuchas50%.
BIG-IPWebAcceleratorDynamicCompression–EnablesBIG-IPWebAcceleratortocompressdynamicdatafromwebapplications.
BIG-IPWebAcceleratorDynamicCompressionisdifferentfromstandardcompressionimplementationsbecauseofitscompressionefficiency,anditsabilitytoavoidwidespreadbrowsercompressionbugs.
Furtherenhancementsareutilizedwhenservingdynamic,unique,ormodifiedrequestsforcompresseddatafromthecache.
EvendynamiccontentrequiringuniquesessionIDswithineverylinkonthepagecanbedeliveredandcompressedwithzerocompressionoverhead.
BIG-IPWebAcceleratorIntelligentBrowserReferencing(IBR)–IBRisagroupofcapabilitiesthateliminatestheneedforthebrowsertodownloadrepetitiveorduplicatedata,aswellasensuresthebestuseofbandwidthbycontrollingbrowserbehavior.
Byreducingtheextraconditionalrequestsandexcessdata(re-)transmittedbetweenthebrowserandthewebapplication,IBRreducestheeffectsofWANlatencyanderrors.
IBRalsosignificantlyreducestheamountofdatadownloadedwithoutrequiringJavaappletsormakingchangestothebrowserthatarecommonindeltacompressionmethodologies.
IBRiscomprisedofthreemainfunctionalities–MultiConnect,DynamicContentControl,andDynamicLinearization.
MultiConnect–EnablesInternetExplorertoopenmoresimultaneousconnectionsbetweenthebrowserandwebapplication,allowingincreasedparalleldatatransfers.
MultiConnectisextremelyeffectiveonhighlatency/highbandwidthnetworkssuchassatelliteandmobilenetworks.
DynamicContentControl(DCC)–Eliminatesthedownloadofrepetitivedatabyensuringthatthebrowserdownloadsonlythedatathatistrulydynamicandunique.
DCCalsoeliminatesbrowser'conditionalrequests'forstaticdatathatisincorrectlyconsidereddynamicwhileensuringtrulydynamicanduniquecontentisfreshlyserved.
DynamicLinearization–ServesupindividualpagesofAdobePDFdocumentsfromlargenon-linearPDFfiles,allowingforfastfirst-pageviewsofPDFdocuments.
Onlythepagesthatauserisreadingaretransferred;usersnolongerhavetowaitforanentiremanual,customerform,designspecification,ordrawingtobeloadedpriortoviewing.
SymmetricoperationBydeployingBIG-IPLTMwithWebAcceleratorinasymmetricfashion,userswillrequesttheSharePointdatafromtheirlocal(orclosest)BIG-IP.
IfthelocalBIG-IPdoesnothavetherequesteddata,itwillpullitfromtheBIG-IPthatislocaltotheSharePointfarm.
UsersrequestingcontentfromtheirbranchofficeBIG-IPLTMwithWebAcceleratorwillbenefitfromthesametechnologiesasfortheasymmetricdeployments,aswellas:SharePointDataProximity–CommonlyrequestedSharePointpagesandobjectswillbecachedatthebranchofficeBIG-IPLTMwithWebAcceleratorappliances,greatlyreducingtheneedtore-pullthedataeverytimeitisrequested.
13BIG-IP-to-BIG-IPsecuredatatransferacceleration–SincetheBIG-IPLTMwithWebAcceleratorappliancebecomesbothendpointsoftheTCPconnection,itisabletoleveragehardwareandsoftwarethatisdesignedtotransferdatafasterthannormalbrowsertoserverconnections.
PerformanceresultsThefollowingsectionsdescribethetestsperformed,resultsobserved,andalsoprovideasummaryofkeyfindingsandconclusions.
TestmethodologyThemethodologyemployedwasdesignedtodeterminetheperformanceimprovementsthatcouldberealizedbydeployingtheF5BIG-IPLTMwithWebAcceleratorsolutiontotwoscenarios,comparedtoanun-acceleratedbaseline.
Morethan100individualtestswereperformed,withcomprehensivemonitoringofallserverandnetworkresourceutilization.
Thefollowingwerethekeymethodologystepsintestingbothscenarios:1.
ConfigurethescenariowithoutBIG-IPLTMwithWebAcceleratorappliances(un-acceleratedbaselineconfiguration).
2.
ApplytheappropriateworkloadatincreasingloadlevelsuntilaverageWANutilizationreachedapproximately75%ofWANbandwidth(bestachievableun-acceleratedresult).
3.
RepeatStep2forallfiveWAN'types'(seeTable1,below)toprovideallbaselinedata.
4.
Deploybest-practiceBIG-IPLTMwithWebAcceleratorconfigurationforthescenario.
5.
RepeattestloadachievedinStep2(bestun-acceleratedload)todetermineeffectofBIG-IPLTMwithWebAcceleratoratthatloadlevel(directcomparisonresult).
6.
Continueincreasingloaduntil75%WANutilizationonceagainachieved(bestachievableacceleratedresult).
7.
RepeatSteps5and6forallWAN'types'toprovideall'accelerated'data.
KeyperformancemetricsmonitoredincludedClient(users)observedKBytes/secnetworktraffic,WANnetworktraffic(MbpsorKbps),SharePointrequestspersecond(RPS),HTTPHits/sec,.
NETRequests/sec,WFECPUUtilization%,etc.
Thesedataprovidedacomprehensiveviewof'causeandeffect,'anddemonstratedtheeffectsofBIG-IPLTMwithWebAcceleratoronnetworktraffic(LANandWAN),WFECPU,andachievablethroughput(RPS).
Theresultspresentedbelowforbothscenariosareaverysmallsubsetofallthoserecorded.
TheyareintendedtosummarizetheresultsofSteps2,5,and6(un-acceleratedbaseline,acceleratedcomparisonbaseline,andacceleratedbest-achievable).
Asanaidforyourunderstandingandanalysisoftheresults,afurthersection(Keyfindings)willalsosummarizethekeyobservationsandconclusionsfoundfromtheresults.
WorkloadscenariosTwoworkloadscenarioswereused,eachincludingarangeandmixofcommonSharePointfunctionsthatusersinthescenarioswouldtypicallyandfrequentlyemploy.
TheseincludedHomepage,listbrowsing,search,documentlibrary(doclib)browsing,opening/readingdocuments,documentcheck-out/-in,etc.
OfficeSharePointServer2007isafeature-richandhighlycustomizableapplication,andthuscustomers'workloadsmayincludefunctionality14notincludedinthetestworkloads.
Theworkloadsthatwereused,however,wereintendedtobebroadlyrepresentative.
Inshort–'yourmileagemayvary.
'BranchofficeTheemulatedbranchofficeusersareauthenticatedontotheSharePointsite,andarebothconsumersandsuppliersofinformation(thatis,read/writeactivitiesthatmodifydatacontent).
Thefollowingshowsthefunctionmixused.
Thepercentagesdonotsumto100,butareratherthepercentagelikelihoodofagivenfunctionoccurringforeachpassoftheworkload'script.
'Thescriptloopscontinuouslyduringthetestduration.
Home70%ReadfromEventscalendarlist65%ReadfromAnnouncementslist65%Search(rangeofkeywords)25%–Open/readdocumentfromSearchresultsset10%PickrandomDoclibfolderandbrowsecontent65%–Open/readdocumentfromfolder10%PickspecificDoclibfolderandbrowsecontent10…40%–Check-out/modify/check-indocumentfromfolder10%InternetTheemulatedInternetusershaveanonymousaccesstotheSharePointsite,andareconsumersofinformation(thatis,read-only).
Thefollowingshowsthefunctionmixused.
Asnoread/writeactivitiesareperformed,andmixandpercentagesoftheotherfunctionsareadjustedtosuit(forexample,moresearchingandbrowsing).
Thescriptloopscontinuouslyduringthetestduration.
Home70%ReadfromEventscalendarlist65%ReadfromAnnouncementslist65%Search(rangeofkeywords)40%–Open/readdocumentfromSearchresultsset25%PickrandomDoclibfolderandbrowsecontent65%–Open/readdocumentfromfolder25%NetworkconfigurationsandparametersThesolutions(bothsymmetricandasymmetric)weretestedagainstarangeofemulatedWANpropertiesasregardsbandwidth,latencyandpacketlosspercentage.
ThesewerechosentorepresentarangeoftypicalWANenvironmentsthatwillbeseenglobally.
Table1detailsthetestedWANenvironments,theircharacteristicsandexamplesofreal-worldapplication.
15Table1.
WANNetworkparametersWANTypeBandwidthLatency(mSec)PacketLoss%1.
Regionalconnection(T3)44Mbps90ms0.
1%2.
U.
S.
branchoffice6Mbps90ms0.
1%3.
U.
S.
–EMEA6Mbps200ms0.
1%4.
U.
S.
–EMEAsatellite6Mbps300ms0.
1%5.
U.
S.
–APAC(T1)1536Kbps300ms1.
0%ThefirstWANtypeisaT3connectionat44Mbpsasmightbefoundconnectingcorporateregionalcenters.
Averagelatency,duetodistance,is90ms.
Asmallpercentageofpacketlosswilloccur.
Type2representsaU.
S.
branchofficeconnectionat6Mbps.
Type3isaconnectionbetweenlocationsintheU.
S.
andEurope,withanincreasedlatencyduetothedistance.
Type4issimilar,butrepresentsasatellite(versuscable)connectionwithincreasedlatency.
Finally,type5representsaT1U.
S.
toAsia-Pacificconnectionatlowerbandwidth,alatencyreflectinglongercabledistanceandahigherpacketlosspercentage.
DatatablemetricsTables2and3detailcomparativetestresultsandpresentarangeofkeyperformancemetrics.
Thefollowingdescribesthetablecolumnheadingsandmetrics,toassistyouinunderstandingofthedata.
Load–arelativemeasureofclientloadappliedtothetestsystem.
Notdirectlyrelatedto'supportedusers,'whichcanbebetterderivedfrom'AvgRPS'(seebelow).
WANtype–refertoTable1fordetailsofthefiveWANconfigurationstested.
Thelettersuffixindicatestheresultfor'U'(bestun-acceleratedresult),'D'(accelerateddirectcomparisonresult),and'A'(bestpossibleacceleratedresult).
Forexample,comparingresults2Uand2DshowsthedifferenceobservedwhendeployingBIG-IPLTMwithWebAcceleratorbutrunningthesameload(thatis,reservecapacitygained)foraWANrunningat6Mbps,90mslatencyand0.
1%packetloss.
Whereas,comparing2Uand2AshowstheactualgaininperformancepossiblewithBIG-IPLTMwithWebAccelerator(higherloadandthroughput),bothshownatthemaximumrecommendedWANutilizationrate.
AvgRPS–averageSharePointrequests/second.
A'request'representsauserclickingonanitemresultinginanewpagebeingpresented.
Thisisthekeyperformancethroughputmetric,directlyrelatedtouserperceivedresponsetimes.
Responsetimeislargelyinfluencedbynetworklatency,asWFEservercapacityismorethanadequateinallthesetests.
RPSisalsobroadlyequivalenttoHTTPpages/secandto.
NETrequests/sec.
Avghits/sec–generally,anHTTPpagewillrequiremultipleHTTPHitstorender.
Thiswillvarydependingonpagecomplexity(numberofSharePointwebparts,etc.
)ObservationshowsthattypicalSharePointpageshaveanun-acceleratedHits-to-Pageratioof3:1.
16AvgclientKB/sec–theaverageKilobytes/second(KB/sec)acrossallemulatedclientsasreportedbytheclientemulationtools.
ThisequatestotheUserLANtrafficrate.
AvgPage–theaverageresponsetime(seconds)ofallworkloadfunctionpages,asreportedbytheclientemulationtool.
Statisticallyan'averageofaverages'itprovidesanindicationofoverallresponsetimechangesduetoBIG-IPLTMwithWebAccelerator.
AvgSearch–theaverageresponsetime(seconds)fortheSearchfunctiontoreturnaresultssetpagetotheuser.
Arangeofdifferentsearchkeywordswasused,thusthemin/maxresponsetimeforthisvaries.
AvgListOpen–theaverageresponsetime(seconds)fortheusertobepresentedwithadoclibfoldercontentlistofdocumentsafterclickingopenadoclibfolder.
Multiplefolderswithdifferingcontentwereused.
AvgDocOpen–theaverageresponsetimewhenauserclicksonadocumentinadocliblisttoopen/readit.
Asdifferingsizes/typesofMicrosoftOfficedocumentswereused,thetimewillvarydependingonthesize,availablebandwidthandlatency.
BranchofficeresultsTable2detailsresultsobtainedfortheBranchOfficescenario,bothun-acceleratedandwhenemployingaBIG-IPLTMwithWebAcceleratorsymmetricdeploymentsolution.
Comparativedataisshownfortheun-accelerated(U),directcomparison(D),andaccelerated(A)results.
Table2.
SymmetricconfigurationresultsLoadWANtestAvgRPSAvgHits/secAvgClientKB/sAvgPageAvgSearchAvgListOpenAvgDocOpen3501U40.
00121.
0032052.
952.
045.
818.
673501D56.
7053.
3018970.
440.
492.
080.
656001A93.
7088.
2031110.
590.
922.
690.
78552U6.
2019.
105071.
122.
146.
109.
57552D9.
008.
503000.
350.
4-02.
000.
553002A47.
6044.
7016190.
500.
662.
580.
61703U6.
2019.
105061.
974.
519.
9717.
41703D10.
7010.
003620.
730.
763.
900.
883003A44.
9042.
2015110.
881.
004.
700.
9217LoadWANtestAvgRPSAvgHits/secAvgClientKB/sAvgPageAvgSearchAvgListOpenAvgDocOpen804U6.
1018.
804982.
835.
0413.
5324.
45804D11.
6010.
903871.
111.
165.
811.
003004A42.
5039.
8014301.
261.
426.
391.
12225U1.
534.
701243.
126.
4615.
5426.
70225D3.
122.
971051.
291.
496.
661.
031005A13.
6012.
804621.
531.
727.
671.
19InternetresultsTable3detailsresultsobtainedfortheInternetscenario,bothun-acceleratedandwhenemployingaBIG-IPLTMwithWebAcceleratorasymmetricdeploymentsolution.
Comparativedataisshownfortheun-accelerated(U)andaccelerated(A)results.
Table3.
AsymmetricconfigurationresultsLoadWANtestAvgRPSAvgHits/secAvgClientKB/sAvgPageAvgSearchAvgListOpenAvgDocOpen4001U43.
40127.
4037223.
152.
105.
5410.
824001A56.
4065.
1025231.
041.
203.
325.
55552U4.
9114.
304175.
163.
399.
6119.
42552A7.
839.
013560.
960.
973.
186.
06703U4.
8814.
204158.
255.
1914.
8632.
01703A8.
8210.
103941.
901.
896.
2812.
40804U4.
7013.
4040910.
907.
3118.
9248.
20804A9.
2010.
504022.
712.
588.
7818.
07245U1.
293.
6510812.
708.
2422.
5354.
10245A2.
362.
73974.
164.
6012.
4521.
9018KeyfindingsThefollowingarekeyobservationsfromthedatapresented.
BranchOffice–symmetricBIG-IPLTMwithWebAcceleratorsolutionResultsforthe6Mbpsand1536KbpstestsshowthatdeployingBIG-IPLTMwithWebAcceleratorprovidedalmosteighttimes(8X)thethroughput(RPS)atthesame75%WANtrafficcapacity.
This8XthroughputincreasewasalsoachievedwithanincreaseinClientLANtrafficofonlythreetimes,thusbothWANandClientLANtrafficweresignificantlyoptimized.
Availableemulatedclienttesthardwaredidnotpermitdrivingthe44Mbpsacceleratedteststofull75%WANcapacity;howevertheresultsshowthesametrendsasforthe6Mbpstests,andsimilarpercentagegainsshouldbeexpected.
Thehits-per-pageratioof3:1(un-accelerated)wasreducedtolessthan1:1.
ThisshowssignificantoptimizationinHTTPprotocol,andtheeffectsofclient-sidelocalcachingofpages,contentandfiles.
ThetestsiteHomepage,forexample,wasalwaysreadfromtheBIG-IPLTMwithWebAcceleratorcacheasitscontentdidnotchange.
Averagepage(response)timesimproveddramatically,evenatfullacceleratedload.
Individualfunctionresponsetimesshowedmarkedimprovements,evenatfullacceleratedload.
Searchdecreasedbyafactorof4(minimalpagecontentchange),Folderlistopenbyafactorof2(moderatelist/pagecontentchange),anddocumentopenbyafactorofupto20(majorityofdocumentscachedbyBIG-IPLTMwithWebAccelerator).
TheeffectsofdifferentWANlatency(90msvs.
200msvs.
300ms)werelargelyobviated,especiallyobservingthosefunctionsthatincludedhigherpage/data/filetransfers.
Althoughnotcitedintheresults,areductioninWFECPUutilizationwasalsonotedwhenBIG-IPLTMwithWebAcceleratorwasemployed.
ThisispartlyduetoprotocolefficienciesprovidedbyBIG-IPLTM,andinpartduetoclient-sideBIG-IPWebAcceleratorcaching(reducingtheneedfortheclientbrowsertogototheWFEforpages/files).
ThepositiveeffectsofthevarioustechnologiesBIG-IPLTMwithWebAcceleratoremployed(TCP/IPoptimization,HTTPoptimization,compression,caching,etc.
)canbeclearlyseenacrossthedata.
Internet–asymmetricBIG-IPLTMwithWebAcceleratorsolutionWhilenotabletoprovideallthecapabilitiesandbenefitsofasymmetricsolution,resultsshowthatdeployingasingleBIG-IPLTMwithWebAcceleratorapplianceinthisscenariowillhavemeasurablebenefitsintermsofincreasedthroughputandinprovidinguserswithanimprovedexperience.
Typicalthroughputimprovementsforthe6Mbpsand1536Kbpstestsapproachedafactorof2(doublethethroughput).
Again,itwasnotpossibletoemulatesufficientloadforthe44MbpsteststodrivetheacceleratedWANtocapacity,butthetrendsaresimilartothe6Mbpstestsandthesamedegreeofimprovementshouldbeexpected.
Thehits-per-pageratiodroppedfrom3:1(un-accelerated)toabout1.
2:1(accelerated)showingagooddegreeofprotocoloptimization.
Averagepage(response)timesshowedimprovementsrangingfromfactorsofbetween3and5(thatis,somefunctionstookonefifthofthetime).
ClientLANtrafficwasreducedto75%oftheun-acceleratedcases.
19Agoodlevelofcompressionwasachieved,butnotethattheusers'browsersareusedtoun-compressthedataandneedtobesettodoso.
SummaryThetwoexamplescenariospresentedarebelievedtobebroadlyapplicabletoOfficeSharePointServer2007deployments.
Theyareespeciallyrelevanttocustomersconsideringconsolidatinggeographicallydistributedfarmsintoafewcentralizeddatacenters,andareconcernedthataWANmayprovidereducedperformance.
Theun-acceleratedtestsresultsclearlyshowthepotentialimpactofbothlowerbandwidth(comparedtoaLAN)andtheperformancedegradationthathigherlatencyvaluesandincreasedpacketlosscancause.
ThecomparableresultsachievedwhendeployingaBIG-IPLTMwithWebAcceleratorsolutiondemonstratethatasignificantperformanceimprovementcanbeachievedintermsofincreasedthroughput,reducedresponsetimes,betterWFEefficiencyandoverallnetworkusageoptimization.
TheBIG-IPLTMwithWebAcceleratorapplianceisstraightforwardtoinstallandconfigure,howeverHPrecommendssolicitingassistancefromHPServices(NetworkSystemsGroup)and/orfromF5NetworksInc.
todeterminetheappropriateBIG-IPmodelstosuityourneeds,andforassistanceinthedetailedconfigurationoftheappliances.
HPisgratefulfortheon-sitetechnicalassistanceprovidedbyF5NetworksInc.
duringthelabsetuprequiredforthetestingdocumentedinthispaper.
HPServicesandF5solutionsOneofHP'sstrengthsistobuildallianceswithkeypartnersthataddvaluetothesolutionsweproposetoourcustomers.
Today,HPbelievesthatF5Networksisprovidingourcustomerswithamongthebestvalue-addsolutionsintheApplicationDeliveryController(ADC)arena.
Asaresult,HPhasdevelopedastrongpartnershipwithF5andhasalreadydevelopedaprovenworld-wideADCsolutionsimplementationexperience.
HP'sselectionofF5asoneofitsmainADCpartnersisstronglyreinforcedbytheADCGartnerMagicQuadrant,inwhichF5ispositionedinthefarupperright(source:MagicQuadrantforApplicationDeliveryProducts,2007).
F5andHPhaveacooperativeCustomerSupportAgreementtofacilitatethedeliveryofbest-in-classtechnicalexpertiseandrapidresponseforHPcustomers'supportneeds.
F5andHPareproudtoworktogethertoprovidecustomerswiththebestpossibleuserexperience.
ThetermsoftheagreementenableHPtoresellF5'scompletefamilyofADCsolutionsandtohavethemintegratedandsupportedwithHPnetworkingservices.
Asaresultofthisagreement,HPcustomerscanobtaincompletenetworkdesign,implementation,andsupportservicesfortheirF5products.
HP'sGlobalServiceandSupportcompletesthepicturebyprovidingexpertnetworkingservicesthathelpensureavailabilityofthecustomer'snetworkanywhere,anytimethebusinessrequires.
AcontactURLforHPServices(NetworkSystemsGroup)isprovidedinthesectionFormoreinformation.
FurtherinformationrelatingtoF5BIG-IPsolutionsandproductscanalsobeobtainedfromF5NetworksInc.
perthecontactinformationshowninthesectionFormoreinformation.
20Implementingaproof-of-conceptAsamatterofbestpracticeforalldeployments,HPrecommendsimplementingaproof-of-conceptusingatestenvironmentthatmatchesascloselyaspossibletheplannedproductionenvironment.
Inthisway,appropriateperformanceandscalabilitycharacterizationscanbeobtained.
Forhelpwithaproof-of-concept,contactanHPServicesrepresentative.
Formoreinformation,speakwithanHPsalespersonorreseller,orcontactHPdirectlyusingthecontactinformationavailableathttp://www.
hp.
com/hps/contacts/index.
html21FormoreinformationHPActiveAnswersresourcesforMicrosoftOfficeSharePointServer2007http://www.
hp.
com/solutions/activeanswers/sharepointHPProLiantservershttp://www.
hp.
com/go/proliantHPBladeSystemc-Classservershttp://www.
hp.
com/go/bladesHPStoragetechnologieshttp://www.
hp.
com/go/storageHPServicesNetworkSystemsGrouphttp://www.
hp.
com/go/networksF5NetworksInc.
mainpagehttp://www.
f5.
comF5BIG-IPpagehttp://www.
f5.
com/products/big-ip/F5WebAcceleratorpagehttp://www.
f5.
com/products/big-ip/product-modules/webaccelerator.
htmlF5WhitePaperonIntelligentBrowserReferencinghttp://www.
f5.
com/pdf/white-papers/eliminating-http-chattiness-wp.
pdfF5WhitePaperonTCPoptimizationshttp://www.
f5.
com/pdf/white-papers/tcpexpress-wp.
pdfF5MicrosoftSolutionsandContactshttp://www.
f5.
com/microsoftF5BIG-IPSharePoint2007ApplicationReadyNetworkDeploymentGuidehttp://www.
f5.
com/pdf/deployment-guides/f5-sharepoint07-dg.
pdfTohelpusimproveourdocuments,pleaseprovidefeedbackatwww.
hp.
com/solutions/feedback.
2007Hewlett-PackardDevelopmentCompany,L.
P.
Theinformationcontainedhereinissubjecttochangewithoutnotice.
TheonlywarrantiesforHPproductsandservicesaresetforthintheexpresswarrantystatementsaccompanyingsuchproductsandservices.
Nothinghereinshouldbeconstruedasconstitutinganadditionalwarranty.
HPshallnotbeliablefortechnicaloreditorialerrorsoromissionscontainedherein.
Microsoft,Windows,andSharePointareU.
S.
registeredtrademarksofMicrosoftCorporation.
Intel,ItaniumandXeonaretrademarksofIntelCorporationintheU.
S.
andothercountries.
AMDandAMDOpteronaretrademarksofAdvancedMicroDevices,Inc.
JavaisaUStrademarkofSunMicrosystems,Inc.
AdobeisatrademarkofAdobeSystemsIncorporatedF5,F5Networks,BIG-IP,TMOS,LTM,LocalTrafficManagerandWebAcceleratoraretrademarksorregisteredtrademarksofF5NetworksInc.
4AA1-6598ENW,November2007