In Proceedings of Performance Tools 98 Lecture Notes in Computer Science, Vol. 1468, pp , September 1998

Size: px
Start display at page:

Download "In Proceedings of Performance Tools 98 Lecture Notes in Computer Science, Vol. 1468, pp. 231-242, September 1998"

Transcription

1 In Proceedings of Performance Tools 98 OnChoosingaTaskAssignmentPolicyfora Lecture Notes in Computer Science, Vol. 1468, pp , September 1998 MorHarchol-Balter?;1,MarkE.Crovella??;2,andCristinaD.Murta???;2 DistributedServerSystem 2DepartmentofComputerScience,BostonUniversity 1LaboratoryforComputerScience,MIT Abstract.Weconsideradistributedserversystemmodelandaskwhich policiescommonlyproposedforsuchdistributedserversystems:round- vicedemandisknowninadvance.weconsiderfourtaskassignment hostprocessestasksinfirst-come-first-serveorderandthetask'sser- Robin,Random,Size-Based,inwhichalltaskswithinagivesizerange policyshouldbeusedforassigningtaskstohosts.inourmodeleach areassignedtoaparticularhost,anddynamic-least-work-remaining, criticallyonthevariabilityinthetasksizedistribution.inparticular decisionofwhichtaskassignmentpolicyisbest.wendthatnooneof inwhichataskisassignedtothehostwiththeleastoutstandingwork. wendthatwhenthetasksizesarenothighlyvariable,thedynamic theabovetaskassignmentpoliciesisbestandthattheanswerdepends Ourgoalistounderstandtheinuenceoftasksizevariabilityonthe policyispreferable.howeverwhentasksizesshowthedegreeofvariabilitymorecharacteristicofempiricallymeasuredcomputerworkloads,the Size-Basedpolicyisthebestchoice.Weusetheresultingobservations performthedynamicpolicybyalmost2ordersofmagnitudeandcan outperformothertaskassignmentpoliciesbymanyordersofmagnitude, toargueinfavorofaspecicsize-basedpolicy,sita-e,thatcanout- 1Introduction underarealistictasksizedistribution. ofthistrendincludedistributedwebservers,distributeddatabaseservers,and distributeddesignsbecauseoftheirscalabilityandcost-eectiveness.examples Tobuildhigh-capacityserversystems,developersareincreasinglyturningto highperformancecomputingclusters.insuchasystem,requestsforservicearrive assigningtaskstohostmachinesisknownasthetaskassignmentpolicy. andmustbeassignedtooneofthehostmachinesforprocessing.therulefor???supportedbyagrantfromcapes,brazil.permanentaddress:depto.deinformatica,universidadefederaldoparana,curitiba,pr81531,brazil.??supportedinpartbynsfgrantsccr andccr ?supportedbythensfpostdoctoralfellowshipinthemathematicalsciences.

2 systeminwhicheachincomingtaskisimmediatelyassignedtoahostmachine, andeachhostmachineprocessesitsassignedtasksinrst-come-rst-served (FCFS)order.Wealsoassumethatthetask'sservicedemandisknowninadvance.Ourmotivationforconsideringthismodelisthatitisanabstractionof Inthispaperweconcentrateontheparticularmodelofadistributedserver someexistingdistributedservers,describedinsection3. tributedserversystems:round-robin,inwhichtasksareassignedtohostsina cyclicalfashion;random,inwhicheachtaskisassignedtoeachhostwithequal probability;size-based,inwhichalltaskswithinacertainsizerangearesent toanindividualhost;anddynamic(alsoknownasleast-work-remaining)in Weconsiderfourtaskassignmentpoliciescommonlyproposedforsuchdis- whichanincomingtaskisassignedtothehostwiththeleastamountofoutstandingworklefttodo(basedonthesumofthesizesofthosetasksinthe queue). ofwhichtaskassignmentpolicyisbest.wearemotivatedinthisrespectby inmanymeasurementsofcomputerworkloads.inparticular,measurementsof theincreasingevidenceforhighvariabilityintasksizedistributions,witnessed manycomputerworkloadshavebeenshowntotaheavy-taileddistributions Ourgoalistostudytheinuenceoftasksizevariabilityonthedecision analysisoranalyticapproximations.weshowthatthevariabilityofthetasksize withveryhighvariance,asdescribedinsection2.2. distributionmakesacrucialdierenceinchoosingataskassignmentpolicy,and thatworkswellunderconditionsofhightasksizevariance. weusetheresultingobservationstoargueforaspecictaskassignmentpolicy Incomparingtaskassignmentpolicies,wemakeuseofsimulationsandalso 2BackgroundandPreviousWork ied,butmanybasicquestionsremainopen.inthecasewheretasksizesare 2.1FundamentalResultsinTaskAssignment Theproblemoftaskassignmentinamodellikeourshasbeenextensivelystudtion,theoptimalityofShortest-Linetaskassignmentpolicy(sendthetaskto unknown,thefollowingresultsexist:underanexponentialtasksizedistribu- thehostwiththeshortestqueue)wasprovenbywinston[14]andextended byweber[12]toincludetasksizedistributionswithnondecreasingfailurerate. TheactualperformanceoftheShortest-Linepolicyisnotknownexactly,butis approximatedbynelsonandphillips[9].infactasthevariabilityofthetask alentoptimalityandperformanceresultshavenotbeendevelopedforthetask sizedistributiongrows,theshortest-linepolicyisnolongeroptimal,whitt[13]. assignmentproblem,tothebestofourknowledge.forthescenarioinwhichthe agesofthetaskscurrentlyservingareknown,weber[12]hasshownthatthe Shortest-Expected-Delayruleisoptimalfortasksizedistributionswithincreasingfailurerate,andWhitt[13]hasshownthatthereexisttasksizedistributions Inthecasewheretheindividualtasksizesareknown,asinourmodel,equiv- forwhichtheshortest-expected-delayruleisnotoptimal.

3 1 Distribution of process lifetimes (log plot) (fraction of processes with duration > T) 1/2 1/4 1/8 Fig.1.MeasureddistributionofUNIXprocessCPUlifetimes,from[5].Dataindicates 1/16 fractionofjobswhosecpusevicedemandsexceedtseconds,asafunctionoft. 1/32 1/64 2.2Measurementsoftasksizedistributionsincomputer applications Duration (T secs.) Manyapplicationenvironmentsshowamixtureoftasksizesspanningmanyordersofmagnitude.Insuchenvironmentstherearetypicallymanysmalltasks, andfewerlargetasks.muchpreviousworkhasusedtheexponentialdistribution poormodelandthataheavy-taileddistributionismoreaccurate.ingenerala heavy-taileddistributionisoneforwhichprfx>xgx?;where0<<2. surementsindicatethatformanyapplicationstheexponentialdistributionisa tocapturethisvariability,asdescribedinsection2.1.however,recentmea- 1.Decreasingfailurerate:Inparticular,thelongerataskhasrun,thelonger Tasksizesfollowingaheavy-taileddistributionshowthefollowingproperties: 3.Thepropertythataverysmallfraction(<1%)oftheverylargesttasks 2.Innitevariance(andif1,innitemean). makeupalargefraction(half)oftheload.wewillrefertothisimportant itisexpectedtocontinuerunning. Thelowertheparameter,themorevariablethedistribution,andthemore tasksthatcomprisehalftheload. pronouncedistheheavy-tailedproperty,i.e.thesmallerthefactionoflarge propertythroughoutthepaperastheheavy-tailedproperty. measureddistributionofcpurequirementsofoveramillionunixprocesses, takenfrompaper[5].thisdistributioncloselytsthecurve Asaconcreteexample,Figure1depictsgraphicallyonalog-logplotthe ronments,includinginstructional,reasearch,andadministrativeenvironments. In[5]itisshownthatthisdistributionispresentinavarietyofcomputingenvi- PrfProcessLifetime>Tg=1=T:

4 Infact,heavy-taileddistributionsappeartotmanyrecentmeasurementsof computingsystems.theseinclude,forexample: {UnixprocessCPUrequirements,measuredatUCBerkeley:1[5]. {UnixprocessCPUrequirementsmeasuredatBellcore:11:25[8]. {SizesoflesstoredinUnixlesystems:[7]. {SizesoflestransferredthroughtheWeb:1:11:3[1,3]. {I/Otimes:[11]. servicerequirements. typicallytendstobecloseto1,whichrepresentsveryhighvariabilityintask Inmostofthesecaseswhereestimatesofweremade,12.Infact, {SizesofFTPtransfersintheInternet::91:1[10]. 3ModelandProblemFormulation Weareconcernedwiththefollowingmodelofadistributedserver.Theserver iscomposedofhhosts,eachwithequalprocessingpower.tasksarrivetothe systemaccordingtoapoissonprocesswithrate.whenataskarrivestothe system,itisinspectedbyadispatcherfacilitywhichassignsittooneofthe tasks. hostsforservice.weassumethedispatcherfacilityknowsthesizeofthetask. ThetasksassignedtoeachhostareservedinFCFSorder,andtasksarenot preemptible.weassumethatprocessingpoweristheonlyresourceusedby ontheirjob'sprocessingdemand.ifthejobexceedsthatdemand,itiskilled. batchdistributedcomputingfacilityatmit'slaboratoryforcomputerscience. Thexolasfacilityhasadispatcherfrontendwhichassignseachjobtooneof Xolasconsistsof4identicalmultiprocessorhosts.Usersspecifyanupperbound Theabovemodelforadistributedserverwasinitiallyinspiredbythexolas thehostsforservice.theuserisgivenanupperboundonthetimetheirjob willhavetowaitinthequeue,basedonthesumofthesizesofthejobsinthat queue.thejobsqueuedateachhostareeachruntocompletioninfcfsorder. terizedbythreeparameters:,theexponentofthepowerlaw;k,thesmallest sult,wemodeltasksizesusingadistributionthatfollowsapowerlaw,buthas possibleobservation;andp,thelargestpossibleobservation.theprobability anupperbound.werefertothisdistributionasaboundedpareto.itischarac- Weassumethattasksizesshowsomemaximum(butlarge)value.Asare- massfunctionfortheboundedparetob(k;p;)isdenedas: f(x)= 1?(k=p)x??1kxp: k andvaryovertherange0to2inordertoobservetheeectofchanging ThroughoutthispaperwemodeltasksizesusingaB(k;p;)distribution, (1) variabilityofthedistribution.tofocusontheeectofchangingvariance,we keepthedistributionalmeanxed(at3000)andthemaximumvaluexed(at

5 10 14 Second Moment of Bounded Pareto Distribution 10 power law f(x) w/ exponent α Fig.2.ParametersoftheBoundedParetoDistribution(left);SecondMomentof B(k;1010;)asafunctionof,whenEfXg=3000(right). 9 Numberofhosts Systemload 8 Meanservicetime h=8. =: k p Taskarrivalprocess EfXg=3000timeunits Poissonwithrate=1=EfXgh=:0021 Maximumtaskservicetime parameter Minimumtaskservicetime p=1010timeunits chosensothatmeantaskservicetimestays 0<2 constantasvaries(0<k1500) tasks/unittime p=1010).inordertokeepthemeanconstant,weadjustkslightlyaschanges Table1.Parametersusedinevaluatingtaskassignmentpolicies thisdistributionwillstillshowveryhighvariabilityifkp.forexample, (0<k1500).TheaboveparametersaresummarizedinTable1. isnotaheavy-taileddistributioninthesensewehavedenedabove.however, Figure2(right)showsthesecondmomentEX2 functionofforp=1010,wherekischosentokeepefxgconstantat3000,(0< NotethattheBoundedParetodistributionhasallitsmomentsnite.Thus,it declines.furthermore,theboundedparetodistributionalsostillexhibitsthe k1500).thegureshowsthatthesecondmomentexplodesexponentiallyas heavy-tailedpropertyand(tosomeextent)thedecreasingfailurerateproperty ofthisdistributionasa oftheunboundedparetodistribution. thebesttaskassignmentpolicy.thefollowingfourarecommonchoices: Random:anincomingtaskissenttohostiwithprobability1=h.Thispolicy Giventheabovemodelofadistributedserversystem,weaskhowtoselect Round-Robin:tasksareassignedtohostsincyclicalfashionwiththeith taskbeingassignedtohostimodh.thispolicyalsoequalizestheexpected equalizestheexpectednumberoftasksateachhost.

6 Size-Based:Eachhostservestaskswhoseservicedemandfallsinadesignated largetasks. range.thispolicyattemptstokeepsmalltasksfromgetting\stuck"behind timesthanrandom. numberoftasksateachhost,andtypicallyhaslessvariabilityininterarrival Dynamic:Eachincomingtaskisassignedtothehostwiththesmallestamount ofoutstandingwork,whichisthesumofthesizesofthetasksinthehost's queueplustheworkremainingonthattaskcurrentlybeingserved.this tasksizes.theeectivenessofthesetaskassignmentschemeswillbemeasured Inthispaperwecomparethesepoliciesasafunctionofthevariabilityof systemstandpointattemptstoachieveinstantaneousloadbalance. policyisoptimalfromthestandpointofanindividualtask,andfroma 3.1ANewSize-BasedTaskAssignmentPolicy:SITA-E intermsofmeanwaitingtimeandmeanslowdown,whereatask'sslowdownis Beforedelvingintosimulationandanalyticresults,weneedtospecifyafew itswaitingtimedividedbyitsservicedemand.allmeansareper-taskaverages. moreparametersofthesize-basedpolicy. taskissenttotheappropriatehostbasedonitssize.inpracticethesizeranges fortasksofsizebetween15minutesand3hours,a6-hourqueue,a12-hour associatedwiththehostsareoftenchosensomewhatarbitrarily.theremightbe a15-minutequeuefortasksofsizebetween0and15minutes,a3-hourqueue Insize-basedtaskassignment,asizerangeisassociatedwitheachhostanda queueandan18-hourqueue,forexample.(thisexampleisusedinpracticeat thecornelltheorycenteribmsp2jobscheduler[6].) ment,whichwerefertoassita-e SizeIntervalTaskAssignmentwithEqual Load.Theideaissimple:denethesizerangeassociatedwitheachhostsuch thatthetotalwork(load)directedtoeachhostisthesame.themotivationfor doingthisisthatbalancingtheloadminimizesmeanwaitingtime. Inthispaperwechooseamoreformalalgorithmforsize-basedtaskassign- iseasytoobtainbymaintainingahistogram(inthedispatcherunit)ofalltask theexpectedworkdirectedtoeachhostisthesame.thetasksizedistribution sizeswitnessedoveraperiodoftime. thetasksizedistributiontodenethecutopoints(deningtheranges)sothat Themechanismforachievingbalancedexpectedloadatthehostsistouse ofhosts.thenwedetermine\cutopoints"xi,i=0:::hwherek=x0<x1< p(possiblyequaltoinnity)denotethelargesttasksize,andhbethenumber functionoftasksizeswithnitemeanm.letkdenotethesmallesttasksize, x2<:::<xh?1<xh=p,suchthat Moreprecisely,letF(x)=PrfXxgdenotethecumulativedistribution andassigntotheithhostalltasksranginginsizefromxi?1toxi. Zx1 x0=kxdf(x)=zx2 x1xdf(x)==zxh=p xh?1xdf(x)=mh=rpkxdf(x) h

7 10 8 Simulated mean waiting time 10 5 Simulated mean slowdown 10 7 Random. Round Robin 10 4 Random... Dynamic. Round Robin 10 6 SITA E Dynamic SITA E mean.intheremainderofthepaperwewillalwaysassumethetasksizedistributionistheboundedparetodistribution,b(k;p;). SITA-Easdenedcanbeappliedtoanytasksizedistributionwithnite policiesviasimulation.simulationparametersareasshownintable1. 4SimulationResults InthissectionwecomparetheRandom,Round-Robin,SITA-E,andDynamic outsetwellbelowthetruemean;thetruemeanisn'tachieveduntilenoughlarge below[2].thisoccursbecausetherunningaverageoftasksizesistypicallyatthe tasksarrive.theconsequenceforasystemlikeourownisthatsimulationoutputs dicultbecausethesystemapproachessteadystateveryslowlyandusuallyfrom Simulatingaserversystemwithheavy-tailed,highlyvariableservicetimesis appearmoreoptimisticthantheywouldinsteady-state.tomakeoursimulation only.eachdatapointshowninourplotsistheaverageof400independentruns, measurementslesssensitivetothestartuptransient,werunoursimulationfor 4105arrivalsandthencapturedatafromthenextsinglearrivaltothesystem eachofwhichstartedfromanemptysystem. ability).asdescribedinsection2.2,valuesintherange1.0to1.3tendtobe commoninempiricalmeasurementsofcomputingsystems. Weconsidervaluesintherange1.1(highvariability)to1.9(lowervari Fig.3.MeanWaitingTime(a)andMeanSlowdown(b)underSimulationofFour 10 1 TaskAssignmentPoliciesasaFunctionof. (b) Figure3showstheperformanceofthesystemforallfourpolicies,asafunctionof(notethelogarithmicscaleontheyaxis).Figure3(a)showsmean waitingtimeand3(b)showsmeanslowdown.belowwesimplysummarizethese results;inthenextsection,wewilluseanalysistoexplaintheseresults.

8 androundrobinpoliciesissimilar,andthatbothcasesperformmuchmore poorlythantheothertwo(sita-eanddynamic).asdeclines,bothofthe performancemetricsundertherandomandround-robinpoliciesexplodeapproximatelyexponentially.thisgivesanindicationofthesevereimpactsthat Firstofall,observethattheperformanceofthesystemundertheRandom heavy-tailedworkloadscanhaveinsystemswithnaivetaskassignmentpolicies. namicisontheorderof100timesbetterforbothmetricswhencomparedto RandomandRoundRobin.Forlarge,thismeansthatDynamicperformsquite well withmeanslowdownlessthan1.howeverasthevariabilityintasksize increases(as!1),dynamicisunabletomaintaingoodperformance.ittoo TheDynamicpolicyshowsthebenetsofinstantaneousloadbalancing.Dy- suersfromroughlyexponentialexplosioninperformancemetricsasdeclines. undersita-eisrelativelyunchanged,withmeanslowdownalwaysbetween2 formancemetricsforrandom,roundrobin,anddynamicallexplode,sita-e's three.overtheentirerangeofvaluesstudied,theperformanceofthesystem and3.thisisthemoststrikingaspectofourdata:inarangeofinwhichper- Incontrast,thebehaviorofSITA-Eisquitedierentfromthatoftheother performanceremainsremarkablyinsensitivetoincreaseintasksizevariability. canbeontheorderof100timesbetterthanthatofdynamic. morecharacteristicofempiricalmeasurements(1:1),sita-e'sperformance mentexhibitsbetterperformance;butwhentasksizesshowthevariabilitythatis Asaresultwendthatwhentasksizeislessvariable,Dynamictaskassign- SITA-EbecomespreferabletoDynamicoveralargerrangeof. acrosstherangeoffrom1.1to1.9isdiculttounderstandusingthetoolsof In[4]wesimulatearangeofloads()andshowthatasloadincreases, simulationalone.forthatreasonthenextsectiondevelopsanalysisofsita-e andtheotherpolicies,andusesthatanalysistoexplainsita-e'sperformance. TheremarkableconsistencyofsystemperformanceundertheSITA-Epolicy Tounderstandthedierencesbetweentheperformanceofthefourtaskassignmentpolicies,weprovideafullanalysisoftheRound-Robin,Random,and 5AnalysisofTaskAssignmentPolicies formulabelowwhichanalyzesthem/g/1fcfsqueue: SITA-Epolicies,andanapproximationoftheDynamicpolicy. IntheanalysisbelowwewillrepeatedlymakeuseofthePollaczek-Kinchin wheredenotestherateofthearrivalprocess,xdenotestheservicetime EfWaitingTimeg=EX2 EfSlowdowng=EfW=Xg=EfWgEX?1 =2(1?) [Pollaczek-Kinchinformula] distribution,anddenotestheutilization(=efxg).theslowdownformulas followfromthefactthatwandxareindependentforafcfsqueue.

9 tailed,thesecondmomentoftheservicetimeexplodes,asshowninfigure2. RandomTaskAssignment.TheRandompolicysimplyperformsBernoullisplittingontheinputstream,withtheresultthateachhostbecomesanindependenerallypoorbecausethesecondmomentoftheB(k;p;)ishigh. metricsareproportionaltothesecondmomentofb(k;p;).performanceisgen- is,i=.sothepollaczek-kinchinformulaappliesdirectly,andallperformance M=B(k;p;)=1queue.Theloadattheithhost,isequaltothesystemload,that thesecondmomentoftheservicetime.recallthatiftheworkloadisheavy- ObservethateverymetricforthesimpleFCFSqueueisdependentonEX2, RoundRobin.TheRoundRobinpolicysplitstheincomingstreamsoeachhost formanceclosetotherandomcasesinceitstillseeshighvariabilityinservice seesaneh=b(k;p;)=1queue,withutilizationi=.thissystemhasper- SITA-E.TheSITA-EpolicyalsoperformsBernoullisplittingonthearrival times,whichdominatesperformance. stream(whichfollowsfromourassumptionthattasksizesareindependent).by thedenitionofsita-e,i=.howeverthetasksizesateachqueuearedeterminedbytheparticularvaluesoftheintervalcutos,fxig;i=0;:::;h.infact, hostiseesam=b(xi?1;xi;)=1queue.thereasonforthisisthatpartitioning oftheresultingregionstounitprobabilityyieldsanewsetofboundedpareto theboundedparetodistributionintocontiguousregionsandrenormalizingeach distributions.in[4]weshowhowtocalculatethesetofxisfortheb(k;p;) distribution,andwepresenttheresultingformulasthatprovidefullanalysisof thesystemunderthesita-epolicyforalltheperformancemetrics. knownapproximationsfortheperformancemetricsofthem/g/hqueue[15]: performedthesimulationstudy.however,in[4]weprovethatadistributed assignmentisactuallyequivalenttoanm/g/hqueue.fortunately,thereexist Dynamic.TheDynamicpolicyisnotanalyticallytractable,whichiswhywe systemofthetypeinthispaperwithhhostswhichperformsdynamictask wherexdenotestheservicetimedistributionandqdenotesthenumberin queue.what'simportanttoobservehereisthatthemeanqueuelength,and thereforethemeanwaitingtimeandmeanslowdown,areallproportionaltothe EQM=G=h =EQM=M=h EX2 =EfXg2; andround-robintaskassignmentpolicies. secondmomentoftheservicetimedistribution,aswasthecasefortherandom icyoverthewholerangeof.figure5againshowstheseanalytically-derived assignmentpoliciesoverarangeofvalues.figure4showstheanalyticallyderivedmeanwaitingtimeandmeanslowdownofthesystemundereachpol- metrics,butonlyovertherangeof12,whichistherangeofcorrespondingtomostempiricalmeasurementsofprocesslifetimesandlesizes(see Section2.2).(Notethat,becauseofslowsimulationconvergenceasdescribedat thebeginningofsection4,simulationvaluesaregenerallylowerthananalytic Usingtheaboveanalysiswecancomputetheperformanceoftheabovetask predictions;howeverallsimulationtrendsagreewithanalysis).

10 boththeseguresgrowsworseasdecreases,wheretheperformancecurves followthesameshapeasthesecondmomentoftheboundedparetodistribution,showninfigure2.thisisexpectedsincetheperformanceofrandomand Dynamicisdirectlyproportionaltothesecondmomentoftheservicetimedistribution.Bycontast,lookingatFigure5weseethatintherange1<<2, themeanwaitingtimeandespeciallymeanslowdownunderthesita-epolicy isremarkablyconstant,withmeanslowdownsaround3,whereasrandomand Dynamicexplodeinthisrange.TheinsensitivityofSITA-E'sperformanceto FirstobservethattheperformanceoftheRandomandDynamicpoliciesin inthisrangeisthemoststrikingpropertyofoursimulationsandanalysis. Analytically derived mean waiting time Random 10 9 SITA E... Dynamic Approximation 10 8 Analytically derived mean slowdown Random SITA E Dynamic Approximation <2. Fig.4.Analysisofmeanwaitingtimeandmeanslowdownoverwholerangeof, Analytically derived mean waiting time: Upclose Random SITA E... Dynamic Approximation Analytically derived mean slowdown: Upclose 10 7 Random 10 6 SITA E... Dynamic Approximation rangeof,12. Fig.5.Analysisofmeanwaitingtimeandmeanslowdownoverempiricallyrelevant

11 adynamicpolicyexplodes?acarefulanalysisoftheperformanceofsita-eat eachqueueofthesystem(see[4])leadsustothefollowinganswers: 1.Bylimitingtherangeoftasksizesateachhost,SITA-Egreatlyreduces WhydoesSITA-Eperformsowellinaregionoftasksizevariabilitywherein 2.Whenloadisbalanced,themajorityoftasksareassignedtothelow-numbered thevarianceofthetasksizedistributionwitnessedbythelowered-numbered hosts,therebyimprovingperformanceatthesehosts.infacttheperformance atmosthostsissuperiortothatofanm/m/1queuewithutilization. 3.Furthermore,meanslowdownisimprovedbecausesmalltasksobserveproportionatelylowerwaitingtimes. hosts,whicharethehostswiththebestperformance.thisisintensiedby theheavy-tailedpropertywhichimpliesthatveryfewtasksareassignedto highnumberedhosts. systemperformanceeventuallydeterioratesbadly.thereasonisthatasoverallvariabilityintasksizesincreases,eventuallyevenhost1willwitnesshigh Forthecaseof1,showninFigure4,evenundertheSITA-Epolicy, overwhichsita-eshowsgoodperformance.forexample,whenthenumberof hostsis32,sita-e'sperformancedoesnotdeteriorateuntil:8. variability.furtheranalysis[4]indicatesthataddinghostscanextendtherange 6Conclusion Dynamic(sendingthetasktothehostwiththeleastremainingwork). inuenceswhichtaskassignmentpolicyisbestinadistributedsystem.we Inthispaperwehavestudiedhowthevariabilityofthetasksizedistribution onthevariabilityoftasksizedistribution.whenthetasksizesarenothighly considerfourpolicies:random,round-robin,sita-e(asize-basedpolicy),and variable,thedynamicpolicyispreferable.however,whentasksizesshowthe SITA-Eisbest. degreeofvariabilitymorecharacteristicofempiricalmeasurements(1), Wendthatthebestchoiceoftaskassignmentpolicydependscritically teristicofempiricalmeasurements,sita-eoutperformsdynamicbycloseto2 severalordersofmagnitude.andintherangeoftasksizevariabilitycharac- large:randomandround-robinareinferiortobothsita-eanddynamicby Themagnitudeofthedierenceinperformanceofthesepoliciescanbequite fourpoliciesgleanedfromouranalysis: theirperformanceisdirectlyproportionaltothesecondmomentofthetask OuranalysisoftheRandom,Round-RobinandDynamicpoliciesshowsthat Moreimportantthantheaboveresults,though,istheinsightsaboutthese sizedistribution,whichexplainswhytheirperformancedeterioratesasthetask sizevariabilityincreases.thus,eventhedynamicpolicy,whichcomesclosesto achievinginstantaneousloadbalanceanddirectseachtasktothehostwhere

12 itwaitstheleast,isnotcapableofcompensatingfortheeectofincreasing varianceinthetasksizedistribution. SITA-Epolicywhichisasimpleformalizationofsize-basedpolicies,denedto afullanalysisofthesita-epolicy,leadingtoa3-foldcharacterizationofits equalizetheexpectedloadateachhost.thisformalizationallowsustoobtain power:(i)bylimitingtherangeoftasksizesateachhost,sita-egreatlyreduces Tounderstandwhysize-basedpoliciesaresopowerful,weintroducethe aresenttothesubsetthehostshavingthebestperformance.(iii)meanslowdown thevariabilityofthetasksizedistributionwitnessedbyeachhost{thereby isimprovedbecausesmalltasksobserveproportionatelylowerwaitingtimes. These3propertiesallowSITA-Etoperformverywellinaregionoftasksize improvingtheperformanceatthehost.(ii)whenloadisbalanced,mosttasks variabilityinwhichthedynamicpolicybreaksdown. References 1.M.E.CrovellaandA.Bestavros.Self-similarityinWorldWideWebtrac:Evidenceandpossiblecauses.IEEE/ACMTransactionsonNetworking,5(6):835{846, 3.M.E.Crovella,M.S.Taqqu,andA.Bestavros.Heavy-tailedprobabilitydistributionsintheworldwideweb.InAPracticalGuideToHeavyTails,pages1{23. December M.E.CrovellaandL.Lipsky.Long-lastingtransientconditionsinsimulationswith heavy-tailedworkloads.in1997wintersimulationconference, M.Harchol-Balter,M.E.Crovella,andC.D.Murta.Onchoosingataskassignmentpolicyforadistributedserversystem.TechnicalReportMIT-LCS-TR-757, Chapman&Hall,NewYork, S.Hotovy,D.Schneider,andT.O'Donnell.Analysisoftheearlyworkloadonthe 5.M.Harchol-BalterandA.Downey.Exploitingprocesslifetimedistributionsfor CornellTheoryCenterIBMSP2.TechnicalReport96TR234,CTC,Jan dynamicloadbalancing.acmtransactionsoncomputersystems,15(3),1997. MITLaboratoryforComputerScience, G.Irlam.Unixlesizesurvey. 8.W.E.LelandandT.J.Ott.Load-balancingheuristicsandprocessbehavior.In 10.V.PaxsonandS.Floyd.Wide-areatrac:ThefailureofPoissonmodeling. 9.R.D.NelsonandT.K.Philips.Anapproximationforthemeanresponsetimefor ProceedingsofPerformanceandACMSigmetrics,pages54{69, D.L.PetersonandD.B.Adams.FractalpatternsinDASDI/Otrac.InCMG IEEE/ACMTransactionsonNetworking,pages226{244,June1995. Evaluation,17:123{139,1998. shortestqueueroutingwithgeneralinterarrivalandservicetimes.performance 12.R.W.Weber.Ontheoptimalassignmentofcustomerstoparallelservers.Journal 13.WardWhitt.Decidingwhichqueuetojoin:Somecounterexamples.Operations Proceedings,December1996. ofappliedprobability,15:406{413, R.W.Wol.StochasticModelingandtheTheoryofQueues.PrenticeHall, W.Winston.Optimalityoftheshortestlinediscipline.JournalofAppliedProbability,14:181{189,1977. Research,34(1):226{244,January1986.

How To Choose Between The Two Of The Four Task Assignment Policies In A Distributed Server System

How To Choose Between The Two Of The Four Task Assignment Policies In A Distributed Server System In Proceedings of Performance Tools 98 Lecture Notes in Computer Science, Vol. 1468, pp. 231-242, September 1998 On Choosing a Task Assignment Policy for a Distributed Server System Mor Harchol-Balter?

More information

SIMULATION OF LOAD BALANCING ALGORITHMS: A Comparative Study

SIMULATION OF LOAD BALANCING ALGORITHMS: A Comparative Study SIMULATION OF LOAD BALANCING ALGORITHMS: A Comparative Study Milan E. Soklic Abstract This article introduces a new load balancing algorithm, called diffusive load balancing, and compares its performance

More information

Optimization for QoS on Web-Service-Based Systems with Tasks Deadlines 1

Optimization for QoS on Web-Service-Based Systems with Tasks Deadlines 1 Optimization for QoS on Web-Service-Based Systems with Tasks Deadlines 1 Luís Fernando Orleans Departamento de Engenharia Informática Universidade de Coimbra Coimbra, Portugal lorleans@dei.uc.pt Pedro

More information

Students will benefit from pencils with erasers, if possible since revisions are part of learning.

Students will benefit from pencils with erasers, if possible since revisions are part of learning. Suggestions and/or Directions for Implementing Extended Concept (2) Activities Students will benefit from pencils with erasers, if possible since revisions are part of learning. Students should be allowed

More information

Analyzing Mission Critical Voice over IP Networks. Michael Todd Gardner

Analyzing Mission Critical Voice over IP Networks. Michael Todd Gardner Analyzing Mission Critical Voice over IP Networks Michael Todd Gardner Organization What is Mission Critical Voice? Why Study Mission Critical Voice over IP? Approach to Analyze Mission Critical Voice

More information

Scheduling. Yücel Saygın. These slides are based on your text book and on the slides prepared by Andrew S. Tanenbaum

Scheduling. Yücel Saygın. These slides are based on your text book and on the slides prepared by Andrew S. Tanenbaum Scheduling Yücel Saygın These slides are based on your text book and on the slides prepared by Andrew S. Tanenbaum 1 Scheduling Introduction to Scheduling (1) Bursts of CPU usage alternate with periods

More information

DynamicLoadBalancing ExploitingProcessLifetimeDistributionsfor MorHarchol-Balter and AllenB.Downey tionwhetherpreemptivemigration(migratingactiveprocesses)isnecessary,orwhetherremote WeconsiderpoliciesforCPUloadbalancinginnetworksofworkstations.Weaddresstheques-

More information

Payment minimization and Error-tolerant Resource Allocation for Cloud System Using equally spread current execution load

Payment minimization and Error-tolerant Resource Allocation for Cloud System Using equally spread current execution load Payment minimization and Error-tolerant Resource Allocation for Cloud System Using equally spread current execution load Pooja.B. Jewargi Prof. Jyoti.Patil Department of computer science and engineering,

More information

Stochastic Processes and Queueing Theory used in Cloud Computer Performance Simulations

Stochastic Processes and Queueing Theory used in Cloud Computer Performance Simulations 56 Stochastic Processes and Queueing Theory used in Cloud Computer Performance Simulations Stochastic Processes and Queueing Theory used in Cloud Computer Performance Simulations Florin-Cătălin ENACHE

More information

Operating Systems. III. Scheduling. http://soc.eurecom.fr/os/

Operating Systems. III. Scheduling. http://soc.eurecom.fr/os/ Operating Systems Institut Mines-Telecom III. Scheduling Ludovic Apvrille ludovic.apvrille@telecom-paristech.fr Eurecom, office 470 http://soc.eurecom.fr/os/ Outline Basics of Scheduling Definitions Switching

More information

Lecture Outline Overview of real-time scheduling algorithms Outline relative strengths, weaknesses

Lecture Outline Overview of real-time scheduling algorithms Outline relative strengths, weaknesses Overview of Real-Time Scheduling Embedded Real-Time Software Lecture 3 Lecture Outline Overview of real-time scheduling algorithms Clock-driven Weighted round-robin Priority-driven Dynamic vs. static Deadline

More information

How To Use A Cloud For A Local Cluster

How To Use A Cloud For A Local Cluster Marcos Dias de Assunção 1,2, Alexandre di Costanzo 1 and Rajkumar Buyya 1 1 Department of Computer Science and Software Engineering 2 National ICT Australia (NICTA) Victoria Research Laboratory The University

More information

Process Scheduling CS 241. February 24, 2012. Copyright University of Illinois CS 241 Staff

Process Scheduling CS 241. February 24, 2012. Copyright University of Illinois CS 241 Staff Process Scheduling CS 241 February 24, 2012 Copyright University of Illinois CS 241 Staff 1 Announcements Mid-semester feedback survey (linked off web page) MP4 due Friday (not Tuesday) Midterm Next Tuesday,

More information

IMPROVED LOAD BALANCING MODEL BASED ON PARTITIONING IN CLOUD COMPUTING

IMPROVED LOAD BALANCING MODEL BASED ON PARTITIONING IN CLOUD COMPUTING Available Online at www.ijcsmc.com International Journal of Computer Science and Mobile Computing A Monthly Journal of Computer Science and Information Technology ISSN 2320 088X IJCSMC, Vol. 3, Issue.

More information

Chapter 2: Linear Equations and Inequalities Lecture notes Math 1010

Chapter 2: Linear Equations and Inequalities Lecture notes Math 1010 Section 2.1: Linear Equations Definition of equation An equation is a statement that equates two algebraic expressions. Solving an equation involving a variable means finding all values of the variable

More information

Internet measurements at complexnetworks.fr

Internet measurements at complexnetworks.fr Internet measurements at complexnetworks.fr Guillaume Valadon - http://valadon.complexnetworks.fr LIP6 (CNRS - UPMC) Complex Networks team http://complexnetworks.fr 2 The team http://complexnetworks.fr

More information

How To Model A System

How To Model A System Web Applications Engineering: Performance Analysis: Operational Laws Service Oriented Computing Group, CSE, UNSW Week 11 Material in these Lecture Notes is derived from: Performance by Design: Computer

More information

Load Balancing. Load Balancing 1 / 24

Load Balancing. Load Balancing 1 / 24 Load Balancing Backtracking, branch & bound and alpha-beta pruning: how to assign work to idle processes without much communication? Additionally for alpha-beta pruning: implementing the young-brothers-wait

More information

A Job Self-Scheduling Policy for HPC Infrastructures

A Job Self-Scheduling Policy for HPC Infrastructures A Job Self-Scheduling Policy for HPC Infrastructures F. Guim, J. Corbalan Barcelona Supercomputing Center {francesc.guim,julita.corbalan}@bsc.edu Abstract. The number of distributed high performance computing

More information

Sums of Independent Random Variables

Sums of Independent Random Variables Chapter 7 Sums of Independent Random Variables 7.1 Sums of Discrete Random Variables In this chapter we turn to the important question of determining the distribution of a sum of independent random variables

More information

A Model of Optimum Tariff in Vehicle Fleet Insurance

A Model of Optimum Tariff in Vehicle Fleet Insurance A Model of Optimum Tariff in Vehicle Fleet Insurance. Bouhetala and F.Belhia and R.Salmi Statistics and Probability Department Bp, 3, El-Alia, USTHB, Bab-Ezzouar, Alger Algeria. Summary: An approach about

More information

IP-COM Automatic Call Distributor (ACD)

IP-COM Automatic Call Distributor (ACD) IP-COM Automatic Call Distributor (ACD) The ACD (Automatic Call Distributor) provides the queuing and routing mechanism, receiving and processing requests from incoming calls. Calls are routed to agents

More information

Comparative Analysis of Load Balancing Algorithms in Cloud Computing

Comparative Analysis of Load Balancing Algorithms in Cloud Computing Comparative Analysis of Load Balancing Algorithms in Cloud Computing Ms.NITIKA Computer Science & Engineering, LPU, Phagwara Punjab, India Abstract- Issues with the performance of business applications

More information

NOVEL PRIORITISED EGPRS MEDIUM ACCESS REGIME FOR REDUCED FILE TRANSFER DELAY DURING CONGESTED PERIODS

NOVEL PRIORITISED EGPRS MEDIUM ACCESS REGIME FOR REDUCED FILE TRANSFER DELAY DURING CONGESTED PERIODS NOVEL PRIORITISED EGPRS MEDIUM ACCESS REGIME FOR REDUCED FILE TRANSFER DELAY DURING CONGESTED PERIODS D. Todinca, P. Perry and J. Murphy Dublin City University, Ireland ABSTRACT The goal of this paper

More information

A Task-Based Adaptive-TTL approach for Web Server Load Balancing *

A Task-Based Adaptive-TTL approach for Web Server Load Balancing * A Task-Based Adaptive-TTL approach for Web Server Load Balancing * Devarshi Chatterjee Zahir Tari RMIT University School of Computer Science and IT Melbourne, Australia zahirt@cs cs.rmit.edu.au * Supported

More information

Load Balancing Techniques

Load Balancing Techniques Load Balancing Techniques 1 Lecture Outline Following Topics will be discussed Static Load Balancing Dynamic Load Balancing Mapping for load balancing Minimizing Interaction 2 1 Load Balancing Techniques

More information

Elevator Simulation and Scheduling: Automated Guided Vehicles in a Hospital

Elevator Simulation and Scheduling: Automated Guided Vehicles in a Hospital Elevator Simulation and Scheduling: Automated Guided Vehicles in a Hospital Johan M. M. van Rooij Guest Lecture Utrecht University, 31-03-2015 from x to u The speaker 2 Johan van Rooij - 2011 current:

More information

Road Map. Scheduling. Types of Scheduling. Scheduling. CPU Scheduling. Job Scheduling. Dickinson College Computer Science 354 Spring 2010.

Road Map. Scheduling. Types of Scheduling. Scheduling. CPU Scheduling. Job Scheduling. Dickinson College Computer Science 354 Spring 2010. Road Map Scheduling Dickinson College Computer Science 354 Spring 2010 Past: What an OS is, why we have them, what they do. Base hardware and support for operating systems Process Management Threads Present:

More information

Systems Modelling and Simulation (Lab session 3)

Systems Modelling and Simulation (Lab session 3) Systems Modelling and Simulation (Lab session 3) After this session you should understand. How to model resource failures. 2. How to schedule resources. 3. How to add animations Resource pictures Entity

More information

Fair load-balance on parallel systems for QoS 1

Fair load-balance on parallel systems for QoS 1 Fair load-balance on parallel systems for QoS 1 Luis Fernando Orleans, Pedro Furtado CISUC, Department of Informatic Engineering, University of Coimbra Portugal {lorleans, pnf}@dei.uc.pt Abstract: Many

More information

Justifying Simulation. Why use simulation? Accurate Depiction of Reality. Insightful system evaluations

Justifying Simulation. Why use simulation? Accurate Depiction of Reality. Insightful system evaluations Why use simulation? Accurate Depiction of Reality Anyone can perform a simple analysis manually. However, as the complexity of the analysis increases, so does the need to employ computer-based tools. While

More information

An Approach to Load Balancing In Cloud Computing

An Approach to Load Balancing In Cloud Computing An Approach to Load Balancing In Cloud Computing Radha Ramani Malladi Visiting Faculty, Martins Academy, Bangalore, India ABSTRACT: Cloud computing is a structured model that defines computing services,

More information

Data Science at U of U

Data Science at U of U Data Science at U of U Je M. Phillips Assistant Professor, School of Computing Center for Extreme Data Management, Analysis, and Visualization Director, Data Management and Analysis Track University of

More information

A Practical Guide to Trade Policy Analysis. http://vi.unctad.org/tpa

A Practical Guide to Trade Policy Analysis. http://vi.unctad.org/tpa A Practical Guide to Trade Policy Analysis A joint UNCTAD WTO publication http://vi.unctad.org/tpa Table of contents 1. Analyzing trade flows 2. Quantifying trade policy 3. Analyzing bilateral trade using

More information

SCHEDULING IN CLOUD COMPUTING

SCHEDULING IN CLOUD COMPUTING SCHEDULING IN CLOUD COMPUTING Lipsa Tripathy, Rasmi Ranjan Patra CSA,CPGS,OUAT,Bhubaneswar,Odisha Abstract Cloud computing is an emerging technology. It process huge amount of data so scheduling mechanism

More information

Solving Exponential Equations

Solving Exponential Equations Solving Exponential Equations Deciding How to Solve Exponential Equations When asked to solve an exponential equation such as x + 6 = or x = 18, the first thing we need to do is to decide which way is

More information

Deciding which process to run. (Deciding which thread to run) Deciding how long the chosen process can run

Deciding which process to run. (Deciding which thread to run) Deciding how long the chosen process can run SFWR ENG 3BB4 Software Design 3 Concurrent System Design 2 SFWR ENG 3BB4 Software Design 3 Concurrent System Design 11.8 10 CPU Scheduling Chapter 11 CPU Scheduling Policies Deciding which process to run

More information

Multiobjective Multicast Routing Algorithm

Multiobjective Multicast Routing Algorithm Multiobjective Multicast Routing Algorithm Jorge Crichigno, Benjamín Barán P. O. Box 9 - National University of Asunción Asunción Paraguay. Tel/Fax: (+9-) 89 {jcrichigno, bbaran}@cnc.una.py http://www.una.py

More information

MA 1125 Lecture 14 - Expected Values. Friday, February 28, 2014. Objectives: Introduce expected values.

MA 1125 Lecture 14 - Expected Values. Friday, February 28, 2014. Objectives: Introduce expected values. MA 5 Lecture 4 - Expected Values Friday, February 2, 24. Objectives: Introduce expected values.. Means, Variances, and Standard Deviations of Probability Distributions Two classes ago, we computed the

More information

Waiting Times Chapter 7

Waiting Times Chapter 7 Waiting Times Chapter 7 1 Learning Objectives Interarrival and Service Times and their variability Obtaining the average time spent in the queue Pooling of server capacities Priority rules Where are the

More information

VOIP TRAFFIC SHAPING ANALYSES IN METROPOLITAN AREA NETWORKS. Rossitza Goleva, Mariya Goleva, Dimitar Atamian, Tashko Nikolov, Kostadin Golev

VOIP TRAFFIC SHAPING ANALYSES IN METROPOLITAN AREA NETWORKS. Rossitza Goleva, Mariya Goleva, Dimitar Atamian, Tashko Nikolov, Kostadin Golev International Journal "Information Technologies and Knowledge" Vol.2 / 28 181 VOIP TRAFFIC SHAPING ANALYSES IN METROPOLITAN AREA NETWORKS Rossitza Goleva, Mariya Goleva, Dimitar Atamian, Tashko Nikolov,

More information

Basic Multiplexing models. Computer Networks - Vassilis Tsaoussidis

Basic Multiplexing models. Computer Networks - Vassilis Tsaoussidis Basic Multiplexing models? Supermarket?? Computer Networks - Vassilis Tsaoussidis Schedule Where does statistical multiplexing differ from TDM and FDM Why are buffers necessary - what is their tradeoff,

More information

Teo Chung Piaw National University of Singapore. Copyright 2010, SAS Institute Inc. All rights reserved.

Teo Chung Piaw National University of Singapore. Copyright 2010, SAS Institute Inc. All rights reserved. Teo Chung Piaw National University of Singapore The Business Analytics Movement: Implications for Research and Education The Emergence of Business Analytics MODEL DATA STATISTICS CMPUTER SCIENCE OPERATIONS

More information

Efficient DNS based Load Balancing for Bursty Web Application Traffic

Efficient DNS based Load Balancing for Bursty Web Application Traffic ISSN Volume 1, No.1, September October 2012 International Journal of Science the and Internet. Applied However, Information this trend leads Technology to sudden burst of Available Online at http://warse.org/pdfs/ijmcis01112012.pdf

More information

Financial Time Series Analysis (FTSA) Lecture 1: Introduction

Financial Time Series Analysis (FTSA) Lecture 1: Introduction Financial Time Series Analysis (FTSA) Lecture 1: Introduction Brief History of Time Series Analysis Statistical analysis of time series data (Yule, 1927) v/s forecasting (even longer). Forecasting is often

More information

Observingtheeffectof TCP congestion controlon networktraffic

Observingtheeffectof TCP congestion controlon networktraffic Observingtheeffectof TCP congestion controlon networktraffic YongminChoi 1 andjohna.silvester ElectricalEngineering-SystemsDept. UniversityofSouthernCalifornia LosAngeles,CA90089-2565 {yongminc,silvester}@usc.edu

More information

RESOURCE ALLOCATION FOR INTERACTIVE TRAFFIC CLASS OVER GPRS

RESOURCE ALLOCATION FOR INTERACTIVE TRAFFIC CLASS OVER GPRS RESOURCE ALLOCATION FOR INTERACTIVE TRAFFIC CLASS OVER GPRS Edward Nowicki and John Murphy 1 ABSTRACT The General Packet Radio Service (GPRS) is a new bearer service for GSM that greatly simplify wireless

More information

A Load Balancing Model Based on Cloud Partitioning for the Public Cloud

A Load Balancing Model Based on Cloud Partitioning for the Public Cloud IEEE TRANSACTIONS ON CLOUD COMPUTING YEAR 2013 A Load Balancing Model Based on Cloud Partitioning for the Public Cloud Gaochao Xu, Junjie Pang, and Xiaodong Fu Abstract: Load balancing in the cloud computing

More information

International Journal of Innovative Research in Science, Engineering and Technology Vol. 2, Issue 5, May 2013

International Journal of Innovative Research in Science, Engineering and Technology Vol. 2, Issue 5, May 2013 ISSN: 2319-8753 International Journal of Innovative Research in Science, Engineering and Technology Vol. 2, Issue 5, May 2013 of vibration are 0.14 rad/s and 0.42 rad/s respectively. The dynamic response

More information

International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 575 ISSN 2229-5518

International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 575 ISSN 2229-5518 International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 575 Simulation-Based Approaches For Evaluating Load Balancing In Cloud Computing With Most Significant Broker Policy

More information

Webpage: www.ijaret.org Volume 3, Issue XI, Nov. 2015 ISSN 2320-6802

Webpage: www.ijaret.org Volume 3, Issue XI, Nov. 2015 ISSN 2320-6802 An Effective VM scheduling using Hybrid Throttled algorithm for handling resource starvation in Heterogeneous Cloud Environment Er. Navdeep Kaur 1 Er. Pooja Nagpal 2 Dr.Vinay Guatum 3 1 M.Tech Student,

More information

Graph models for the Web and the Internet. Elias Koutsoupias University of Athens and UCLA. Crete, July 2003

Graph models for the Web and the Internet. Elias Koutsoupias University of Athens and UCLA. Crete, July 2003 Graph models for the Web and the Internet Elias Koutsoupias University of Athens and UCLA Crete, July 2003 Outline of the lecture Small world phenomenon The shape of the Web graph Searching and navigation

More information

Important Notice: In case of divergent interpretation, the original Turkish text shall prevail.

Important Notice: In case of divergent interpretation, the original Turkish text shall prevail. Unofficial Translation of Communique on the Implementation of Annex-4 of the By-Law on Quality of Service in the Electronic Communications Sector prepared by Information and Communication Technologies

More information

Example 1: Dear Abby. Stat Camp for the Full-time MBA Program

Example 1: Dear Abby. Stat Camp for the Full-time MBA Program Stat Camp for the Full-time MBA Program Daniel Solow Lecture 4 The Normal Distribution and the Central Limit Theorem 188 Example 1: Dear Abby You wrote that a woman is pregnant for 266 days. Who said so?

More information

International Journal of Computer Science Trends and Technology (IJCST) Volume 2 Issue 4, July-Aug 2014

International Journal of Computer Science Trends and Technology (IJCST) Volume 2 Issue 4, July-Aug 2014 RESEARCH ARTICLE An Efficient Priority Based Load Balancing Algorithm for Cloud Environment Harmandeep Singh Brar 1, Vivek Thapar 2 Research Scholar 1, Assistant Professor 2, Department of Computer Science

More information

Is Truck Queuing Productive? Study of truck & shovel operations productivity using simulation platform MineDES

Is Truck Queuing Productive? Study of truck & shovel operations productivity using simulation platform MineDES Is Truck Queuing Productive? Study of truck & shovel operations productivity using simulation platform MineDES Dmitry Kostyuk Specialist Scientist, Group Resource and Business Optimisation 25 November

More information

COMMUTATION INSTRUCTIONS

COMMUTATION INSTRUCTIONS COMMUTATION INSTRUCTIONS The following examples illustrate various methods of commuting permanent disability and life pension benefits. Examples A, B and C apply to permanent disability and utilize Table

More information

Modeling and Simulation of Queuing Scheduling Disciplines on Packet Delivery for Next Generation Internet Streaming Applications

Modeling and Simulation of Queuing Scheduling Disciplines on Packet Delivery for Next Generation Internet Streaming Applications Modeling and Simulation of Queuing Scheduling Disciplines on Packet Delivery for Next Generation Internet Streaming Applications Sarhan M. Musa Mahamadou Tembely Matthew N. O. Sadiku Pamela H. Obiomon

More information

Sample Solutions for Assignment 2.

Sample Solutions for Assignment 2. AMath 383, Autumn 01 Sample Solutions for Assignment. Reading: Chs. -3. 1. Exercise 4 of Chapter. Please note that there is a typo in the formula in part c: An exponent of 1 is missing. It should say 4

More information

Prof. Olivier de Weck

Prof. Olivier de Weck ESD.36 System Project Management + Lecture 9 Probabilistic Scheduling Instructor(s) Prof. Olivier de Weck Dr. James Lyneis October 4, 2012 Today s Agenda Probabilistic Task Times PERT (Program Evaluation

More information

Load Balancing & Termination

Load Balancing & Termination Load Load & Termination Lecture 7 Load and Termination Detection Load Load Want all processors to operate continuously to minimize time to completion. Load balancing determines what work will be done by

More information

OPCFailover in Action

OPCFailover in Action Explanation through Example Overview OPC server monitoring should be simple and at times can appear to be. It is easy for this simplicity to lead us into a false sense of security and can lead to being

More information

LBPerf: An Open Toolkit to Empirically Evaluate the Quality of Service of Middleware Load Balancing Services

LBPerf: An Open Toolkit to Empirically Evaluate the Quality of Service of Middleware Load Balancing Services LBPerf: An Open Toolkit to Empirically Evaluate the Quality of Service of Middleware Load Balancing Services Ossama Othman Jaiganesh Balasubramanian Dr. Douglas C. Schmidt {jai, ossama, schmidt}@dre.vanderbilt.edu

More information

Pr(X = x) = f(x) = λe λx

Pr(X = x) = f(x) = λe λx Old Business - variance/std. dev. of binomial distribution - mid-term (day, policies) - class strategies (problems, etc.) - exponential distributions New Business - Central Limit Theorem, standard error

More information

Fruit Machine. Level. Activity Checklist Follow these INSTRUCTIONS one by one. Test Your Project Click on the green flag to TEST your code

Fruit Machine. Level. Activity Checklist Follow these INSTRUCTIONS one by one. Test Your Project Click on the green flag to TEST your code Introduction: This is a game that has three sprites that change costume. You have to stop them when they re showing the same picture (like a fruit machine!). Activity Checklist Follow these INSTRUCTIONS

More information

A New Extension of the Exponential Distribution

A New Extension of the Exponential Distribution Revista Colombiana de Estadística Junio 2014, volumen 37, no. 1, pp. 25 a 34 A New Extension of the Exponential Distribution Una nueva extensión de la distribución exponencial Yolanda M. Gómez 1,a, Heleno

More information

Takeover Suggestion A Registrar Redundancy Handling Optimization. for Reliable Server Pooling Systems

Takeover Suggestion A Registrar Redundancy Handling Optimization. for Reliable Server Pooling Systems University of Duisburg-Essen, Institute for Experimental Mathematics Takeover Suggestion A Registrar Redundancy Handling Optimization for Reliable Server Pooling Systems Institute for Experimental Mathematics

More information

System Architecture Performance Modeling with SystemC

System Architecture Performance Modeling with SystemC System Architecture Performance Modeling with SystemC 9 th European SystemC Users Group Meeting February 17 th, 2004 Raimar Thudt Tino Bertram Infineon Präsentation November 2003 Seite 1 Copyright Infineon

More information

Chapter 7 - Roots, Radicals, and Complex Numbers

Chapter 7 - Roots, Radicals, and Complex Numbers Math 233 - Spring 2009 Chapter 7 - Roots, Radicals, and Complex Numbers 7.1 Roots and Radicals 7.1.1 Notation and Terminology In the expression x the is called the radical sign. The expression under the

More information

Analysis of Delayed Reservation Scheme in Server-based QoS Management Network

Analysis of Delayed Reservation Scheme in Server-based QoS Management Network Analysis of Delayed Reservation Scheme in Server-based QoS Management Network Takeshi Ikenaga Ý, Kenji Kawahara Ý, Tetsuya Takine Þ, and Yuji Oie Ý Ý Dept. of Computer Science and Electronics, Kyushu Institute

More information

A Load Balancing Algorithm based on the Variation Trend of Entropy in Homogeneous Cluster

A Load Balancing Algorithm based on the Variation Trend of Entropy in Homogeneous Cluster , pp.11-20 http://dx.doi.org/10.14257/ ijgdc.2014.7.2.02 A Load Balancing Algorithm based on the Variation Trend of Entropy in Homogeneous Cluster Kehe Wu 1, Long Chen 2, Shichao Ye 2 and Yi Li 2 1 Beijing

More information

A Multiple Access Protocol for Multimedia Transmission over Wireless Networks

A Multiple Access Protocol for Multimedia Transmission over Wireless Networks A Multiple Access Protocol for Multimedia Transmission over Wireless Networks Hong Yu and Mohammed Arozullah Department of Electrical Engineering and Computer Science Capitol College, Maryland, USA yhong@capitol-college.edu

More information

INSURANCE RISK THEORY (Problems)

INSURANCE RISK THEORY (Problems) INSURANCE RISK THEORY (Problems) 1 Counting random variables 1. (Lack of memory property) Let X be a geometric distributed random variable with parameter p (, 1), (X Ge (p)). Show that for all n, m =,

More information

CloudSimDisk: Energy-Aware Storage Simulation in CloudSim

CloudSimDisk: Energy-Aware Storage Simulation in CloudSim CloudSimDisk: Energy-Aware Storage Simulation in CloudSim Baptiste Louis, Karan Mitra, Saguna Saguna and Christer Åhlund Department of Computer Science, Electrical and Space Engineering Luleå University

More information

Comparative Analysis of Congestion Control Algorithms Using ns-2

Comparative Analysis of Congestion Control Algorithms Using ns-2 www.ijcsi.org 89 Comparative Analysis of Congestion Control Algorithms Using ns-2 Sanjeev Patel 1, P. K. Gupta 2, Arjun Garg 3, Prateek Mehrotra 4 and Manish Chhabra 5 1 Deptt. of Computer Sc. & Engg,

More information

TRAFFIC ENGINEERING OF DISTRIBUTED CALL CENTERS: NOT AS STRAIGHT FORWARD AS IT MAY SEEM. M. J. Fischer D. A. Garbin A. Gharakhanian D. M.

TRAFFIC ENGINEERING OF DISTRIBUTED CALL CENTERS: NOT AS STRAIGHT FORWARD AS IT MAY SEEM. M. J. Fischer D. A. Garbin A. Gharakhanian D. M. TRAFFIC ENGINEERING OF DISTRIBUTED CALL CENTERS: NOT AS STRAIGHT FORWARD AS IT MAY SEEM M. J. Fischer D. A. Garbin A. Gharakhanian D. M. Masi January 1999 Mitretek Systems 7525 Colshire Drive McLean, VA

More information

Real Time Bus Monitoring System by Sharing the Location Using Google Cloud Server Messaging

Real Time Bus Monitoring System by Sharing the Location Using Google Cloud Server Messaging Real Time Bus Monitoring System by Sharing the Location Using Google Cloud Server Messaging Aravind. P, Kalaiarasan.A 2, D. Rajini Girinath 3 PG Student, Dept. of CSE, Anand Institute of Higher Technology,

More information

Modeling and Simulation of a Pacing Engine for Proactive Campaigns in Contact Center Environment

Modeling and Simulation of a Pacing Engine for Proactive Campaigns in Contact Center Environment Modeling and Simulation of a Pacing Engine for Proactive Campaigns in Contact Center Environment Nikolay Korolev, Herbert Ristock, Nikolay Anisimov Genesys Telecommunications Laboratories (an Alcatel-Lucent

More information

ARTIFICIAL NEURAL NETWORKS FOR ADAPTIVE MANAGEMENT TRAFFIC LIGHT OBJECTS AT THE INTERSECTION

ARTIFICIAL NEURAL NETWORKS FOR ADAPTIVE MANAGEMENT TRAFFIC LIGHT OBJECTS AT THE INTERSECTION The 10 th International Conference RELIABILITY and STATISTICS in TRANSPORTATION and COMMUNICATION - 2010 Proceedings of the 10th International Conference Reliability and Statistics in Transportation and

More information

A Hybrid Load Balancing Policy underlying Cloud Computing Environment

A Hybrid Load Balancing Policy underlying Cloud Computing Environment A Hybrid Load Balancing Policy underlying Cloud Computing Environment S.C. WANG, S.C. TSENG, S.S. WANG*, K.Q. YAN* Chaoyang University of Technology 168, Jifeng E. Rd., Wufeng District, Taichung 41349

More information

Optimizing Contact Center Performance with Operational Reports Document Version: 1.0 October 2015

Optimizing Contact Center Performance with Operational Reports Document Version: 1.0 October 2015 Optimizing Contact Center Performance with Operational Reports Document Version: 1.0 October 2015 Introduction LiveEngage is designed to help you attain operational excellence. Agent managers can use LiveEngage

More information

IST 301. Class Exercise: Simulating Business Processes

IST 301. Class Exercise: Simulating Business Processes IST 301 Class Exercise: Simulating Business Processes Learning Objectives: To use simulation to analyze and design business processes. To implement scenario and sensitivity analysis As-Is Process The As-Is

More information

Advanced Operating Systems CS428

Advanced Operating Systems CS428 Advanced Operating Systems CS428 Lecture TEN Semester I, 2009-10 Graham Ellis NUI Galway, Ireland DIY Parallelism MPI is useful for C and Fortran programming. DIY Parallelism MPI is useful for C and Fortran

More information

Call Center Statistics Web Monitor. Call Center Statistics Web Monitor

Call Center Statistics Web Monitor. Call Center Statistics Web Monitor Call Center Statistics Web Monitor Table of Contents Table of Contents...2 Introduction...3 Application Requirements...3 Starting the Call Center Web Monitor...4 The Call Center Web Monitor Screen...5

More information

REAL TIME OPERATING SYSTEMS. Lesson-18:

REAL TIME OPERATING SYSTEMS. Lesson-18: REAL TIME OPERATING SYSTEMS Lesson-18: Round Robin Time Slicing of tasks of equal priorities 1 1. Common scheduling models 2 Common scheduling models Cooperative Scheduling of ready tasks in a circular

More information

UNITY CALL CENTER REPORTING. Version 1.0

UNITY CALL CENTER REPORTING. Version 1.0 UNITY CALL CENTER REPORTING Version 1.0 Contents Introduction... 3 Accessing the Reports... 3 Abandoned Call Report... 4 Settings:... 4 Report:... 5 Agent Activity Detail Report... 6 Settings:... 6 Report:...

More information

Automatic Call Distribution For Business and Call Centers

Automatic Call Distribution For Business and Call Centers Automatic Call Distribution For Business and Call Centers Making Connections Matters Busy organizations require a phone system that can do more than simply handle the traffic. Calls need to get answered

More information

Operating System: Scheduling

Operating System: Scheduling Process Management Operating System: Scheduling OS maintains a data structure for each process called Process Control Block (PCB) Information associated with each PCB: Process state: e.g. ready, or waiting

More information

Introduction. The Supplement shows results of locking using a range of smoothing parameters α, and checkerboard tests.

Introduction. The Supplement shows results of locking using a range of smoothing parameters α, and checkerboard tests. Auxiliary Material Submission for Paper 2014JB010945 Robert McCaffrey Portland State University Inter-seismic locking on the Hikurangi subduction zone: Uncertainties from slow-slip events Introduction

More information

Performance Evaluation of Dynamic Bandwidth Allocation Algorithm for XG-PON Using Traffic Monitoring and Intra- Scheduling

Performance Evaluation of Dynamic Bandwidth Allocation Algorithm for XG-PON Using Traffic Monitoring and Intra- Scheduling , pp. 27-216 http://dx.doi.org/1.14257/ijseia.215.9.8.19 Performance Evaluation of Dynamic Bandwidth Allocation Algorithm for XG-PON Using Traffic Monitoring and Intra- Scheduling Man Soo Han Optical Engineering

More information

A Load Balancing Method in SiCo Hierarchical DHT-based P2P Network

A Load Balancing Method in SiCo Hierarchical DHT-based P2P Network 1 Shuang Kai, 2 Qu Zheng *1, Shuang Kai Beijing University of Posts and Telecommunications, shuangk@bupt.edu.cn 2, Qu Zheng Beijing University of Posts and Telecommunications, buptquzheng@gmail.com Abstract

More information

AN EFFICIENT DISTRIBUTED CONTROL LAW FOR LOAD BALANCING IN CONTENT DELIVERY NETWORKS

AN EFFICIENT DISTRIBUTED CONTROL LAW FOR LOAD BALANCING IN CONTENT DELIVERY NETWORKS Available Online at www.ijcsmc.com International Journal of Computer Science and Mobile Computing A Monthly Journal of Computer Science and Information Technology IJCSMC, Vol. 3, Issue. 9, September 2014,

More information

Operating Systems, 6 th ed. Test Bank Chapter 7

Operating Systems, 6 th ed. Test Bank Chapter 7 True / False Questions: Chapter 7 Memory Management 1. T / F In a multiprogramming system, main memory is divided into multiple sections: one for the operating system (resident monitor, kernel) and one

More information

Modelling Emergence of Money

Modelling Emergence of Money Vol. 117 (2010) ACTA PHYSICA POLONICA A No. 4 Proceedings of the 4th Polish Symposium on Econo- and Sociophysics, Rzeszów, Poland, May 7 9, 2009 Modelling Emergence of Money A.Z. Górski a, S. Drożdż a,b

More information

12.6 Logarithmic and Exponential Equations PREPARING FOR THIS SECTION Before getting started, review the following:

12.6 Logarithmic and Exponential Equations PREPARING FOR THIS SECTION Before getting started, review the following: Section 1.6 Logarithmic and Exponential Equations 811 1.6 Logarithmic and Exponential Equations PREPARING FOR THIS SECTION Before getting started, review the following: Solve Quadratic Equations (Section

More information

Feature Highlights: Automatic Call Distribution (ACD) CTI with E1, T1, PSTN, & IP. Support SS7 Signaling. Screen Popup. Find Last Agent.

Feature Highlights: Automatic Call Distribution (ACD) CTI with E1, T1, PSTN, & IP. Support SS7 Signaling. Screen Popup. Find Last Agent. w w w. g p l e x. c o m Why R Call Center? gplex Call Center allows easy deployment and expansion. Agents can be located remotely at multiple locations while the core server placed at the center office.

More information

Information Theory and Coding Prof. S. N. Merchant Department of Electrical Engineering Indian Institute of Technology, Bombay

Information Theory and Coding Prof. S. N. Merchant Department of Electrical Engineering Indian Institute of Technology, Bombay Information Theory and Coding Prof. S. N. Merchant Department of Electrical Engineering Indian Institute of Technology, Bombay Lecture - 17 Shannon-Fano-Elias Coding and Introduction to Arithmetic Coding

More information

Chapter 1: Order of Operations, Fractions & Percents

Chapter 1: Order of Operations, Fractions & Percents HOSP 1107 (Business Math) Learning Centre Chapter 1: Order of Operations, Fractions & Percents ORDER OF OPERATIONS When finding the value of an expression, the operations must be carried out in a certain

More information

LivePerson Chat Reports

LivePerson Chat Reports LivePerson Chat Reports January 2013 Introduction LivePerson offers a full set of online reports designed to measure the effectiveness of the chat operations of your call center. The statistics provide

More information

Betting with the Kelly Criterion

Betting with the Kelly Criterion Betting with the Kelly Criterion Jane June 2, 2010 Contents 1 Introduction 2 2 Kelly Criterion 2 3 The Stock Market 3 4 Simulations 5 5 Conclusion 8 1 Page 2 of 9 1 Introduction Gambling in all forms,

More information