userpaessler

paessler  时间:2021-03-26  阅读:()
IJCSNSInternationalJournalofComputerScienceandNetworkSecurity,VOL.
7No.
9,September2007103ManuscriptreceivedSeptember5,2007ManuscriptrevisedSeptember20,2007RummageWebServerTuningEvaluationthroughBenchmark(Casestudy:CLICK,andTIMEParameter)HiyamS.
EnsourTheArabAcademyforBankingandFinancialSciences.
Amman,Jordan.
2007.
Dr.
AhmadKayedTheAppliedSciencesUniversity.
Amman,Jordan.
2007.
Abstract-Thisstudyexaminesawebserverperformancetuningbyusingspecialmainparametersinbenchmark,usingrealdataandrealapplicationsinmorethan13differentcases.
Twoadaptiveparameters(CLCIKandTIME)areusedasmeasurementsfortuning.
Awebserverstresstools7benchmark(WSST)isusedasarecognizedapplication.
Someproceduresareprojectedtocomparethefinalresults,thefirstprocessisbasedonfindingthemainfactoroftheparametersaffectingontuning.
Second,avarietyofthevaluesofthebenchmarkparametersarediscussedtohavebetterresultsofthewebserverperformancebyfindingthecorerelationshipamongmainparametersinWSST.
Theparameterscriteriashowtheeffectonwebserverbehaviorundercertainconditionsandenvironments.
Wemonitoritatdifferenttimesandworks.
Contributingdiscusessomeresultssuchas,bottleneck,traffic,andresponsetimewhichrelatedwithcriteria'sandmeasurements.
Keywords:Performance,Webserver,Benchmark,andTuning.
OverviewThispaperpresentstheimportancewebserverperformancetuninginintroductionsectioninfirstsection,andwhyusesbenchmarkasmainsolutionProblemstatementforwebserverisfoundinsection2.
Alltestwebsseverstresstoolsbenchmark(WSST)criteria,thetestenvironment,andmainparameterswillbeshowninsection3.
Observations,scenariosofclickandtimeprocesswillbediscussedinsection4.
Resultsandconclusions,alongwithfuturework.
Willbeaddressedinthelastsection.
1IntroductionTheimportanceofperformancewebserversisquiteclear;therefore,themainpurposeofthisstudyistogainabetterunderstandingofwebserverperformancetuning(WSPtuning).
Webserversdidtaketheperformanceasanintrinsicdesignpremise;thisisacceptableattheearlyadoptionphaseoftheWebserver.
Mostwebserversareusedtoserveasmallgivenloadoverlow-capacitylinks.
Incontrast,nowadays,themainfeaturesofwebserversarestabilizedandcommercialimplementationsarenormal.
Consequently,theimportanceofwebserverperformancetuninghasincreased.
Scalability,reliability,andcontinualityarecrucialelementsinstudyingtheperformancetuning[7,8].
Benchmarksreflecttheperformancebymonitoringtheparametersthatmightaffectthewebserver.
Thisresearchwillstudyawell-knownbenchmarknamedWebServerStressTools7(WSST).
Thefactorstobeusedwillbedefined,andthentheireffectwillbeinvestigatedonawebserverperformanceunderworkloadforacertainapplication.
Thebenchmarkwillbeusedtoevaluatetheperformanceofthewebserverdependingondifferentparameterssuchasusers,delay,time,clicks,ramp,users,URLandrecursivebrowsing.
Software,hardwareandoperatingsystemenvironmentsarefixed.
Weselectonlynaturalfactorsaffectingthewebserverperformance(WSP),whichareCLICK,TIMEandhowtheyarerelatedtoclicktime,clickpersecond,andhitspersecond.
Benchmarkdependsontestingasimulationproceduretorepresentthemodelbehaviorofthewebserverinthetimedomain.
ThesimulatorinbenchmarkrevealsanunpredictedbehavioroftheexaminedWSP.
Thiswouldimplyflexibletechniquesinbenchmarkforperformancetuningevaluation[11,12].
WebServerStressTool(WSST)wasdevelopedbyPaesslerGmbH1[1];itisaconfigurableclient-serverbenchmarkforHTTPserversthatuseworkloadparameters.
Itusesthreeteststomeasuretheserverperformance;namely,HTML,CGI,andAPI.
BysimulatingtheHTTPrequestsgeneratedbymanyusers;i.
e.
;benchmarkcantestWSPundernormalandexcessiveloads[1,4,and5].
Thewebserver(WS)behaviorcanimprovebytuningseveralparameters.
Discoveringthedirectrelationsamongsuchparametersisessentialtodeterminethebestpossiblewebserverbehaviorand,consequently,achieveahighquantitativeperformanceforeachparameterintheWS.
1http://paessler.
netIJCSNSInternationalJournalofComputerScienceandNetworkSecurity,VOL.
7No.
9,September20071042ProblemStatementforWebServerTuningTherearemanywaystotuneawebserver'sperformance.
Theseincludemodeling,analyticalsystem,mathematicalsimulation,andbenchmark.
Benchmarkisusedinthisstudyforanumberofreasons.
Benchmarkgivesusareliable,repeatableandcomparable("standardized")performanceassessment(measurements)ofcompletehardware/softwarewebserverunder(closeto)realisticworkloads[13].
IthasaresponsibilityfortuneWStobestservestaticwebpagesordynamicallycompiledapplicationpages.
Eachwebserverdemandsadifferenthardware,application,andIISperformanceforthetuningoptions.
AnotherconsiderationistheamountoftrafficthatwerealisticallyexpectourWStohandle,particularlyduringthepeakloadperiods.
LoadandtimewillaffecttheWSperformanceandthevaryingbusinesschoices.
Oneshouldbewellacquaintedwithwhattheseloadswillbeandsimulatethemonourserversbeforeputtingthemon-linetoknowhowthewebserverwillperformitsfunction.
Thesearesomereasonswhyitisimportanttorecommendhowtotunethewebserverthroughbenchmark2[15].
2.
1WebServerTuningOneofthedifficultiesintuningthewebserverknowswhattotuneexactlyForthisreason,itisvitaltomonitorthewebservers'behaviorundercertaincriteriaafteradjustingthesettingsofthehardware,software,andwebapplications.
TuningtheWSwillrequireustocarefullymonitorhowchangestoitwillaffecttheperformanceofthewebserver.
First,weshouldknowhowtheserverisfunctioning,andthenwecanmakechangestoimproveperformance.
Changesshouldbemadeonceatatimeandunderanumberofclicks,userswitharollbacktests.
Otherwise,itwillbedifficulttoassesstheimpactofindividualchanges.
Toimprovethewebserverperformancetuning,wewillexamineeverypartoftheWSPparametersofbenchmark.
This,forexample,includestheclicktime,timeforthefirstbyte,timetoconnect,timeforDNS,andtimeforthelocalsocketasmainfactorsthroughthetuningprocess.
2http://microsoft.
com2.
2ProposalSolutionFeedinginformationaboutwebserverhasbeenusedextensivelytosolvemanykindsofWSPproblems.
OneofthefundamentalproprietiesmakingtheseWSPusefulisbenchmarkfortuning.
Inthiswork,weusetwodifferenttypesofwebserverbenchmarkparameters.
Inpreviousstudies,weexaminedallfactorsplayingthemostconspicuouseffectonthebehaviorofthewebserver[15].
Here,however,itisrecommendedtouse(CLICK,TIME)asmainparameterstoguideusinstudyingthewebserver'sbehaviortodealwiththetuningconcept.
2.
3WebServerStressBenchmark(WSST)Performancetestswereusedtoexamineeachpartofthewebserverorthewebapplicationtodiscoverhowtooptimizethemforboostingthewebtraffic(e.
g.
undernumbersofclicks).
WSSTsupportstypesoftestsandiscapableofrunningseveral(e.
g.
20-100)simultaneousrequestsononeURLandrecordtheaveragetimetoprocessthoserequests.
2.
4WhyuseWSSTinourExperimentMostwebsitesandwebapplicationsrunsmoothlyandappropriatelyaslongasonlyoneuserorafewusersarevisitingatthegiventime.
WhathappenswhenthousandsofusersaccessthewebsiteorwebapplicationatthesametimeWhathappenstothewebserverinthiscaseByusingtheWSST,wecansimulatevariousloadpatternsforourwebserver,whichwillhelpusspotproblemsinourwebserverset-up.
Withsteadilyrisingloads(alsocalled"ramptests"),wecanfindouthowmuchloadtheservercanhandlebeforeseriousproblemsarise[1].
TheWSSTcanbeusedforvarioustests[1]:PerformanceTests(PT),LoadTests(LT),StressTests(ST),andRampTests(RT)wherePTareusedtotesteachpartofthewebserverorthewebapplicationtodiscoverhowtobestoptimizethemforhigherwebtraffic.
LTareperformedbytestingthewebsiteusingthebestestimateofthetrafficwebsiteneedstosupport.
Considerthisisa"realworldtest"ofthewebsite.
STconstitutedsimulated"bruteforce"attacksthatapplyexcessiveloadtowebserver.
RTisasetofvariationsofthestresstestsinwhichthenumberofusersraiseduringthetestprocessesfromasingleusertohundredsofusers.
OurtestsneedonlyPT,LT,andST.
IJCSNSInternationalJournalofComputerScienceandNetworkSecurity,VOL.
7No.
9,September20071053TheMainParametersoftheExperimentWehaveadoptedmanytestsusedinliterature[1,2,3,5,and12].
Theyusesometimesalltheparametersatthesametimewithoutbeingspecificandseparate,weindividualtheparametersinourcasejusttotuningourWS.
TheparametersthataretobetakenintoconsiderationinWSSTare:users,clicks,time,delay,ramp,URL,andrecursivebrowsing,thisstudywillfocusonCLICKandTIMEonlywhichhelpstogetaholisticviewofwebsite/webserver/applicationperformance.
WhereCLICKSrepresentfinishtimewheneachuserhasinitiatedagivennumberofclicks.
TIMErepresenttheteststhatrunforaspecifiednumberofminutese.
g.
keepaserverunderfullloadfor15hours.
[1,5]3.
1WSSTParametersExperimentalTestThisBenchmarkingtoolsimulateswebclients,servers,andalargenumberofclient/servertostresswebserver.
Theconfigurationparameterswerefixedinthetestsrunare[1]:Hardwareconfiguration,loadgeneratorsnumberandtype,numberoftherepeating,timeduration,thedelayofclick,runtestwithnumberofclicksperuser,runtestinnumberofminutes,andURLname.
Inourworkwehavesomeconstantsintestsexperimentalasfollows:thenumberofuserare10,weadapt10usersasanormalcase,butbeforewemonitoringthebehaviorsofWSunderworkloadwecheckitunder5,10,and100users,sotheperfectexamplehereisthetestunder10user.
100clickspereveryuseristhebestexampleinourtestthatcomesafterstudyingthenumberofclickperuser.
Werepeatthetests13timesunderdifferentnumbersofclicksandtimeswithchangingtheheterogeneousworkloadthatdoneunder5secondsasconstantofclickdelayinrandomclickdelay,weadapting20MGforeachworkspace.
TheconstantrequirementinWSSTexperimentaltestconfigurationparameterswhichhavefivevariableswithitsvaluesandspecialcommentsinconsecutive:CLICKRunttestfrom5to120clicksperuser,thisistheamountofclickfromthebeginningtotheendoftheWSSTtest.
TIMERuntestfrom5to120perminute,thisistheamountoftimefromthebeginningtotheendofthewebstresstoolstest.
DELAYwith5seconds,howlongatestWSistowaitbeforestartingthetest.
WORKSPACEwith20MB,Thesizeofdata'sfilesusedbyatestWS,eachofdatahasitsownworkspace.
NUMBEROFUSER:with5,10,50,and100.
3.
2TestEnvironmentOurtestsenvironmentspecificationsarefixedeitherinsoftwareorinhardwareasfollows:(CPU,mainMemory,andRAM),ServerSoftware(HTTP),ServerOperatingSystem(windows2000,windowsXP,apacheforwebserver),NetworkSpeedeitherin(Gig,Meg),andthekindofworkload(static,dynamic).
Morespecifically,a64MBofRAMineachclient,a100Base-TXnetworkadapterineachclient,a500MBdiskminimumineachclient,afull-duplex,andswitchednetwork,inServerConfigurationneedCPU:500MHzPentiumIII,RAM:256MB,andNetwork:2x100Base-TX.
[1,2,and7].
3.
3TestWSSTCriteriaAnychanginginclickandtimeparametersinWSSTwillbydefaultmakechanginginsomecriterialikeprotocoltimeforallclicktimes,timeforfirstbyte,timetoconnect,timeforDNS,andtimeforlocal.
Wheretheclicktimerepresentsasimulateduser'smouseclickthatsendsarequest(oneoftheURLsfromtheURLlist)totheserverandimmediatelyrequestinganynecessaryredirects,framesandimages(ifenabled).
Theclicktimeiscalculatedasthetimebetweenwhentheuserclickedandwhentheserverdeliveredtherequestedresourceswithallreferenceditems(imagesetc.
).
AverageClickTimes:showtheaveragevaluesperURL,peruserorperwebsite,TimeforDNStalkedabouttheTimetoresolveaURL'sdomainnameusingtheclientsystem'scurrentDNSserver,alsotheTimetoconnectshowTimetosetupaconnectiontotheserver.
AndthelastcriteriarepresentthetimebetweeninitiatingarequestandreceivingthefirstbyteofdatafromtheserverthatisaTimetofirstbyte(TFB).
3.
4ObservationsThissectiondeterminesbrieflytheWSSTtestscenariosofourexperimentalresearch,whicharebasedonobservationsthataremadeduringthetestingprocess.
3.
4.
1ScenariosofResearchOurprocessesconsistoftwodistinctphases;scenariosdependingontheCLICKparameter,andscenariosdependingontheTIMEparameter.
IJCSNSInternationalJournalofComputerScienceandNetworkSecurity,VOL.
7No.
9,September2007106ProtocolTimesforallURLsUserSimulation:10simultaneoususers-5secondsbetweenclicks(Random)TestType:CLICKS(runtestuntil10clicksperuser)ClickTimeTimetoFirstByteTimetoConnectTimeforDNSTimeforlocalsocketTimeSinceStartofTest[s]20191817161514131211109876543210Time[ms]1701601501401301201101009080706050403020100Figure1.
110clicksperuserinCLICKparameter3.
4.
2CLICKParameterScenario.
Theworkloadofthewebserverispresentedin13stagesrangingfrom5to120clickspersecond.
However,hereweshowtheresultsonlyingraphsthatrepresentcurveactionsinourresearch.
Wewillgiveasampleexampleinthecaseof100clicksperuser.
Thedetailsofresultswillbestatedintheconclusions.
Itisnecessarytoshowgraphsandfinalresultsof10,50,and100clickstovalidatetheargument.
ProtocolTimesforallURLsUserSimulation:10simultaneoususers-5secondsbetweenclicks(Random)TestType:CLICKS(runtestuntil50clicksperuser)ClickTimeTimetoFirstByteTimetoConnectTimeforDNSTimeforlocalsocketTimeSinceStartofTest[s]1101009080706050403020100Time[ms]1401301201101009080706050403020100Figure1.
2(50clicksperuserinCLICKparameter)Figure1describesthecases(10,50,100)intheclickparameter:10clicks:timetofirstbyte,timetoconnect,timeforDNS,andtimeforsocketarerisingslightlybetween0and20ms,buttheclicktimesrisesharplyandthenplummetbetween0and120ms.
50clicks:clicktimesreachthepeakin140msbuttheothercriteriareachaplatedbehaviorwithtimesincethestartoftest(s)between0and150s.
100clicks:clicktimeschangegentlyandrelativelyandtheothercriteriaremainunchangedbutover250mssincestartofthetest.
Wehaveaconspicuouschangecomparedwiththe50clicksintheclickparameter.
Itwasnoticedthattheincreasingnumberofuserswiththehugevolumeofclicksaddstotheworkloadofthewebserver.
Thisdrawsastrongcorrelationbetweentheclickanditscriteria,whicharetheclicktime,timetofirstbyte,timetoconnect,timeforDNS,andtimeforsocket.
ProtocolTimesforallURLsUserSimulation:10simultaneoususers-5secondsbetweenclicks(Random)TestType:CLICKS(runtestuntil100clicksperuser)ClickTimeppppppTimetoFirstByteppppppTimetoConnectppppppTimeforDNSppppppTimeforlocalsocketppppppTimeSinceStartofTest[s]220200180160140120100806040200Time[ms]1601501401301201101009080706050403020100Figure1.
3(100clicksperuserintheCLICKparameter)Figure1:ClickParameters(Clicktime,timeforfirstbyte,timetoconnect,timeforDNS,andtimeforlocalsocket).
3.
4.
3TIMEParameterScenarioTheworkloadofWSispresentedin13stagesfrom5,10,20,to120timespersecond.
However,theresultshereareshowningraphsrepresentingthe10,50,and100timespersecondasasampleonly.
Thecurveactionsrepresentingtheresultswillbeclearintheresultsandconclusionsection.
ProtocolTimesforallURLsUserSimulation:10simultaneoususers-5secondsbetweenclicks(Random)TestType:TIME(runtestfor10minutes)ClickTimeTimetoFirstByteTimetoConnectTimeforDNSTimeforlocalsocketTimeSinceStartofTest[s]550500450400350300250200150100500Time[ms]350300250200150100500Figure2.
110mstimeparameterProtocolTimesforallURLsUserSimulation:10simultaneoususers-5secondsbetweenclicks(Random)TestType:TIME(runtestfor50minutes)ClickTime000000TimetoFirstByte000000TimetoConnect000000TimeforDNS000000Timeforlocalsocket000000TimeSinceStartofTest[s]2,8002,6002,4002,2002,0001,8001,6001,4001,2001,0008006004002000Time[ms]1301201101009080706050403020100Figure2.
250mstimeparameterIJCSNSInternationalJournalofComputerScienceandNetworkSecurity,VOL.
7No.
9,September2007107ProtocolTimesforallURLsUserSimulation:10simultaneoususers-5secondsbetweenclicks(Random)TestType:TIME(runtestfor100minutes)ClickTimeTimetoFirstByteTimetoConnectTimeforDNSTimeforlocalsocketTimeSinceStartofTest[s]5,5005,0004,5004,0003,5003,0002,5002,0001,5001,0005000Time[ms]1009080706050403020100Figure2.
3100mstimeparameterFigure2:Timeparameters(Clicktime,timeforfirstbyte,Timetoconnect,timeforDNS,timeforlocalsocket.
)Figure2describesthecasesof10,50,100msinthetimeparameter:10times:Normalbehaviorswithcriteria(timetofirstbyte,timetoconnect,timeforDNS,andtimeforsocket),exceptforslightchangesintheclicktime.
50times:Theclicktimesincreasesharplyandrelativelywithaconspicuouschangeinthebehaviorofothercriteria(timetofirstbyte,timetoconnect,timeforDNS,andtimeforsocket)comparedwiththeclickparameter.
100times:in2,500stheclicktimesreachthepeakwith100msintimeandastrongdramaticbehavior,andwithaslightsteadystateandarelativechangeinothercriteria.
So,wecandomoreactionsbyextendingthetime.
Itisquiteclearthattheclicktimesinthetimeparameterhaveareversesrelationwiththeclicktimeintheclickparameter.
WSSTshowsthatwecanenhancetheWSbydependingonthetimeparameterwhileraisingthenumberofclicks.
AhighworkloadresultingfromhitsandclickswillnotcauseanyproblemtotheWSifwehaveenoughtimefordoingallthatclicksandhitspersecond.
TheresultperuserandtheresultperURLwillhelpustodosomespecialcalculationslikecountingthenumberofhitsontheWS,andtofindthemaximumandminimumnumberofhitsandK-bitspersecond.
Inaddition,itwillbefeasibletocomparethefinalresultsperURLandperUserfortheCLICKandTIMEparameters,whichcontainssomecriteriasuchasclick,timespent[ms],andaverageclickTime[ms],withtheexistingaverageclicktimeinminutesanddeterminethenumberofusersinourexperimentaltestforallthecasesparameters(click,andtime).
Tables2,3,and3showthisbenefit.
Inthesetwocases(Click,Time),weconcludethatthetimeparameterrisesdramaticallyintheclicktime,whichindicatesthattimeplaysamajorroleinchangingtheWSbehaviors.
Itisbettertoincreasetimewhilewehavemanyclicks,decreasetheloadonWSjustgivenasubmittimeforeveryclick,andstopdoingahundredofclicksorhitsinashortperiodoftime,whichcausesdifficultiesinWSandbadresponses.
Thefirstcolumnintable1and2aredescribesdifferentnumbersofclicks.
Thistellsusthatanincreaseinthenumberofuserswhosendarequest(URL)tothewebserverleadstoanincreaseinthenumberofhitsasacompleteHTTPrequest.
ThistookplaceintheclickparameterinWSST,whichcausedclickduplicationineverysecondandminute,whichmeansanexcessiveloadonthewebserverleadsustohaveanormalresponsetimewithzeroerrorinHTTPrequest.
Consumingthememory,therequestofURL'swithdifferenttypesmakesthewebserversobusy.
Timespent[ms]inthetimeparametersinourtestswithmultipletrialsformorethan13timesindifferentcasesshowsthatthetimespentincreasesinparallelandconcurrencygrowslargerintime.
Dependingonequation1,therearemanydifferentvaluesbetweenthetimespentintimeparametersandthetimespentinclickparametersinordernottowastemuchtime,werecommenddoingmanyrequest(clicks)inashortspanoftimefortheWSwillnotneedopentimestoanswertherequests.
Becausetheserverlosesmuchtimeandmakestheuserwaitforalongtime,wereiterateourrecommendationnottospendmanytimeswithoutmakinggooduse.
Seethesecondcolumnintable3.
Equation1:ThedifferencesbetweenTimeSpent[ms]inCLICK,TIMEparameters.
(1)Ddiffrepresentsthevalueofdifferentfactors.
Themilemeasuresthetimespentsecond,whichisoneofthecriteria.
WhileTIMEandCLICKrepresentthemainIJCSNSInternationalJournalofComputerScienceandNetworkSecurity,VOL.
7No.
9,September2007108parameters,theyareusedinWSST,wherethedotintheequationindicatestheparametertype.
Clicksincreaseintheclickparameterinparallelwiththerisingnumberofclicks.
However,thiswouldbeamassiveincreaseinthetimeparametercomparedwiththesamenumberofclicksundertheclickparameter.
Thetimespent[ms]increasesdirectlywithtimeinthetimeparametermorethanitdoesintheclickparameter.
TheAvg.
clicktime[ms]dropswithtimeinthetimeparametercomparingwiththeclickparameter.
Inotherwords,wehavethehighestvalueintheclickandtimespent[ms]criteriaandthelowestvalueintheAvg.
clicktime[ms]intimeparameter.
Forusers,theaveragetimesingeneralarenormalvaluesiftheaverageiscalculatedwithinalongspanoftime.
Theresults,however,willnotbesatisfactoryifcalculatedfewerthanhundredsofclicks.
(Seetable3)4DiscussionandResultsInthisworkthepurposeofwebserverevaluationsprocessesbyusingWSST,whichisforimprovingtheperformanceandcatchingthemomentoftuninginit.
WhereprotocoltimeforallURLsinallcases(TIME,CLICK)representanHTTPrequestconsistsofseveralstages.
First,theWSnamehastoberesolvedintoanIPaddressusingDNS(TimeforDNS),andthenanIPportisopenedontheserverbytheclienttosendtherequestheader(TimetoConnect).
Theserverthenanswerstherequest(TimetoFirstByte)andsendsalldata.
Whenalldataistransferred,therequestisfinished(ClickTime).
Alsointheabovegraphsalineisshownforthe"timeforlocalsocket"whichisthetimethatWSSTneededtoacquireanopensocketfromtheIPstackofthemachineitrunson.
Forexample,inausualtest,thisvalueshouldalwaysbeinthelowermillisecondarea(1-30ms).
Forextremetraffictests,thisvaluecanriseabove50-100mswhichisasignthattheperformancelimitsofthelocalmachinehavebeenreached,thatwasindicatedanddisplayedinourgraphs.
Dependingontheobservationsabove,weseethatCLICKandTIMEarestronglyrelatedandhaveanimpactontheWStuningevaluation.
IgnoringtheroleofbenchmarkonWSwillcausepoorWSP.
Ifthenumberofclicksislowasshowninourtest(10,50,100clicksperuser),theserverwouldberespondingtorequestsquickly.
Ifthenumberofclicksishigh,respondingtoarequestwillbeslow,becausewewouldhavededicatedtoomuchmemorytothecaches.
Inthiscase,wesuggesttuningtheWSSTtoleaveenoughmemoryfortherestoftheWS.
WealsoneedtoincreasetheamountofRAMonthewebserver,althoughloweringthecachesizescanbeeffective.
Theincreasenumberofclickswouldcausetheworkloadonthewebservertorisedramatically.
Thiswouldsuddenlycausearelativechangetotheresponsetime,increasingthetimegivenforactions,andallowingforfasterresponseswithfewererrorsintheWSP.
Highvolumeoftraffic,whichdependsonthenumberofclicksandhits,makesthememoryloaded.
Aftermonitoringthewebserver,wewonderiftheserverhasenoughmemorysizeornot.
WerecommendthattheminimumamountofRAMneededforthewebserveris128MB,but256MBto1GBwillbebetterfortheWSPtuning.
WeknowthatwemayhaveaproblemwhenWStrafficishighbutthenumberofrequestsbarelybudges.
Whenthathappens,it'slikelythatthereisabottleneckintheWS.
Bottlenecksoccurwiththeriseofthenumberofclicksandperiodsoftimesarelongerthantheyshouldbe.
Weseethatthetimeforthefirstbyte,andothercriteriahavenearlythesamevaluesandbehaviors,exceptforthecriteriaoftheclicktime,whichhasdifferentvaluesandbehaviorsintheclickparameters(Seetable1,2).
However,theyalsohavedifferentvaluesandbehaviorsatthetimeparameters.
Thisshowsthatwecanhaveariseinthetimeconnect,timeforDNS,andlocalsocketwhenthereisachangeinthetimeparameter,becausethebottleneckoftheWSgrowssmaller.
5ConclusionsAllcriteriaforCLICKandTIMEparametersaremeasured,bythat,wehavetodecideifwereducetheserverloadthroughincreasingthetime,anddecreasetheloadsonWS(reverserelation)happensthroughdecreasingthenumbersofclicksandhits,thismakesWSPmoretunableincriteria'sespeciallyonclient'slatency,thatleadustoreducenetworkbandwidthconsumptioneasily,thentheWSPtuningbecomesmorereliablebydefaultifauserhasenoughtimetheyshouldnotworryabouthowmanyclickstheyhadandwhethertheWSisbusyornot.
Becauseuserscandowhatevertheylikewithoutproblemsorerrors,theyshouldjustgivetheserverthetimewhichwebserverneeds.
Weconcludethatifusersdonothavetimeandneedtodotheirworkveryquickly;theyshouldpushthemselvestodecreasethenumberofclicksthatIJCSNSInternationalJournalofComputerScienceandNetworkSecurity,VOL.
7No.
9,September2007109supportthefocusofWSPtuning,makingthewebserverfaster,andmoreefficient.
Wedon'tneedtowaituntiltrafficischokingtheWS,orforcingtoimplementload-balancingsolutionsandthrowingmoreserversattheproblem.
Distributionandobjectarchitectureshelpustoimplementloadbalancingandfaulttolerance.
Load-balancingproductstypicallyarenotrequireduntilaWSscalessohighthattheWSbecomesabottleneckoncethathappensusershavetwochoices:loadbalance,orincreasethebandwidthoftheirconnectionstotheWeb.
Ourparametersareaffecteddirectlyonitcase,soweneedtobemorecarefulwhendetermininghowmuchnumberofclicksandhowlongtimesareavailable3.
SometimesasysteminWSdesignedforacertainleveloftrafficwillspiralintounacceptableresponsetimeswhentrafficincreasesbeyondacertainpoint.
Thisisknownasascalabilityissue.
Weneedachancetoeventuallyencounterabottleneck.
TolocatethebottleneckthatcomesfromraisingthenumberofClickwithspecifictime,weneedtouseaseriesofperformancemonitors.
Thesemonitorsallowuserstoviewtheserverloadandresponsetimeunderavarietyofreal-worldortestconditions.
Responsetimerepresentsthetime(oftenanaverage)thatelapsesbetweentheinitialrequestforinformationandwhenthatdataisdelivered(ornotdelivered,whentheservercan'tprovideitbeforethetimeoutlimitisreached).
WhentheWSisprocessingalargenumberofrequests(underload),itmaytakelongertimetocompletethaniftheserverwereunloaded.
Foruserrequests,thiscanresultinincreasedresponsetimeforclients.
Iftheserverisunderanexcessiveload,dependingonWSSTanalysisweclosetoward"self-tuning"4conceptwhenusebenchmarkasaguideandmaindirectedforWS.
6FutureworkFutureworkwillincludemonitoringthemainparametersinbenchmarkforevaluatingwebserverunderworkloadwithanothercriteria,suchastherelationbetweenClick/hits/users/error/URLatthesametimetuningevaluatethewebserverperformance.
3http://informationweek.
com4http://newsandtech.
com7References[1]http://paessler.
com[2]JohnDilley,"WebServerWorkloadCharacterization",Hewlett-PackardLaboratories.
[3]J.
Dilley,R.
Friedrich,T.
Jin,J.
Rolia.
MeasurementToolsandModelingTechniquesforEvaluatingWebServerPerformance.
HPL-TR-96-161,December1996.
SubmittedtoPerformanceTools'97.
[4]Levy,R.
,etal.
PerformanceManagementforClusterBasedWebServices.
InThe8thIFIP/IEEEInternationalSymposiumonIntegratedNetworkManagement(IM2003).
2003.
ColoradoSprings,Colorado,USA.
[5]Li,C.
,etal.
PerformanceGuaranteeforCluster-BasedInternetServices.
InThe23rdIEEEInternationalConferenceonDistributedComputingSystems(ICDCS2003).
2003.
Providence,RhodeIsland.
[6]Wolf,J.
andP.
S.
Yu,OnBalancingtheLoadinaClusteredWebFarm.
ACMTransactionsonInternetTechnology,2001.
1(2):p.
231-261.
[7]Tapus,C.
,I.
-H.
ChungandJ.
K.
Hollingsworth.
ActiveHarmony:TowardsAutomatedPerformanceTuning.
InSC'02.
2002.
Baltimore,Maryland.
[8]CarlosMaltzahn,KathyJ.
Richardson,andDirkGrunwald.
Performanceissuesofenterpriselevelwebproxies.
InProceedingsoftheACMSigmetricsConferenceonMeasurementandModelingofComputerSystems,Seattle,WA,June1997.
ACM.
[9]JussaraM.
Almeida,VirgilioAlmeida,andDavidJ.
Yates.
MeasuringthebehaviorofaWorld-WideWebserver.
InSeventhConferenceonHighPerformanceNetworking(HPN),pages57–72,WhitePlains,NY,April1997.
IFIP.
[10]M.
Aron,D.
Sanders,P.
Druschel,andW.
Zwaenepoel.
ScalableContent-awareRequestDistributioninCluster-basedNetworkServers.
InProceedingsofthe2000AnnualUSENIXtechnicalConference,SanDiego,CA,June2000.
[11]V.
V.
PanteleenkoandV.
W.
Freeh.
InstantaneousOffloadingofTransientWebServerLoad.
InProceedingsoftheSixthInternationalWorkshoponWebCachingandContentDistribution,Boston,2001.
[12]P.
Joubert,R.
B.
King,R.
Neves,M.
Russinovich,J.
M.
Tracey.
High-PerformanceMemory-BasedWebServers:KernelandUser-SpacePerformance.
InProceedingsof2001USENIXAnnualTechnicalConference,June2001.
[13]StandardPerformanceEvaluationCorporation(SPEC),http://performance.
netlib.
org[14]Riska,A.
,etal.
ADAPTLOAD:EffectiveBalancinginCusteredWebServersUnderTransientLoadIJCSNSInternationalJournalofComputerScienceandNetworkSecurity,VOL.
7No.
9,September2007110Conditions.
In22ndInternationalConferenceonDistributedComputingSystems(ICDCS'02).
2002.
[15]Ribler,R.
L.
,H.
Simitci,andD.
A.
Reed,theAutopilotPerformance-DirectedAdaptiveControlSystem.
FutureGenerationComputerSystems,specialissue(PerformanceDataMining),2001.
18(1):p.
175-187.
Aboutauthors:HiyamS.
Ensour,PHDinCIS(ComputerInformationSystem)fromtheArabAcademyforBankingandFinancialSciences.
Jordan.
MasterinIT(InformationSystem)andBsc.
InComputerSciencefromprincesssumayauniversityfortechnology/RoyalScientificSociety(RSS),Jordan.
WorkinIrbidprivateuniversityaslecturer.
Hayammn@hotmail.
com,hayammn@maktoob.
com.
Dr.
AhmadKayed,theAppliedSciencesUniversity,Kayed_a@asu.
edu.
jo,formoredetailspleasevisit:http://www.
asu.
edu.
jo.

妮妮云(43元/月 ) 香港 8核8G 43元/月 美国 8核8G

妮妮云的来历妮妮云是 789 陈总 张总 三方共同投资建立的网站 本着“良心 便宜 稳定”的初衷 为小白用户避免被坑妮妮云的市场定位妮妮云主要代理市场稳定速度的云服务器产品,避免新手购买云服务器的时候众多商家不知道如何选择,妮妮云就帮你选择好了产品,无需承担购买风险,不用担心出现被跑路 被诈骗的情况。妮妮云的售后保证妮妮云退款 通过于合作商的友好协商,云服务器提供2天内全额退款,超过2天不退款 物...

美国G口/香港CTG/美国T级超防云/湖北高防云服务器物理机促销活动 六一云

六一云 成立于2018年,归属于西安六一网络科技有限公司,是一家国内正规持有IDC ISP CDN IRCS电信经营许可证书的老牌商家。大陆持证公司受大陆各部门监管不好用支持退款退现,再也不怕被割韭菜了!主要业务有:国内高防云,美国高防云,美国cera大带宽,香港CTG,香港沙田CN2,海外站群服务,物理机,宿母鸡等,另外也诚招代理欢迎咨询。官网www.61cloud.net最新直销劲爆...

腾讯云2核4GB内存8M带宽 年74元

一般大厂都是通过首年才有可以享受爆款活动,然后吸引我们注册他们商家达到持续续费和购买的目的。一般只有大厂才能有这样的魄力和能力首年亏本,但是对于一般的公司和个人厂家确实难过,这几年确实看到不少的同类商家难以生存。这里我们可以看到有对应的套餐方案。不过这两个套餐都是100%CPU独享的,不是有某云商家限制CPU的。但是轻量服务器有个不好的就是带宽是较大且流量是限制的额,分别是1GB和1.2TB月流量...

paessler为你推荐
psbc.com95580是什么诈骗信息不点网址就安全吧!seo优化工具SEO优化工具哪个好用点啊?百度关键词分析关键词怎么分析?www.522av.com跪求 我的三个母亲高清在线观看地址 我的三个母亲高清QVOD下载播放地址 我的三个母亲高清迅雷高速下载地址www.765.com有没好的学习网站www.zjs.com.cn请问宅急送客服电话号码是多少?m.kan84.net那里有免费的电影看?杨丽晓博客杨丽晓是如何进入娱乐圈的?sesehu.comwww.hu338.com 怎么看不到啊广告法请问违反了广告法,罚款的标准是什么
广州主机租用 好看的桌面背景图 一点优惠网 标准机柜尺寸 三拼域名 gg广告 域名转接 idc是什么 91vps 双12 彩虹云 无限流量 沈阳主机托管 starry 华为云建站 lamp架构 腾讯数据库 博客域名 privatetracker 腾讯服务器 更多