文件名称：Springer Analyzing Computer Systems Performance with Perl
文件大小：6.95MB
文件格式：PDF
更新时间：2011-03-12 03:57:52
Analyzing Performance Perl Springer Systems
Springer Analyzing Computer Systems Performance with Perl Preface ........................................................ vii Part I Theory of System Performance Analysis 1 Time—The Zeroth Performance Metric.................... 3 1.1 Introduction.......................................... 3 1.2 WhatIsTime?........................................ 4 1.2.1 PhysicalTime................................. 5 1.2.2 SynchronizationandCausality .................. 5 1.2.3 DiscreteandContinuousTime .................. 6 1.2.4 TimeScales................................... 6 1.3 WhatIsaClock?...................................... 8 1.3.1 PhysicalClocks ............................... 8 1.3.2 DistributedPhysicalClocks..................... 9 1.3.3 DistributedProcessing ......................... 9 1.3.4 Binary Precedence. ............................ 10 1.3.5 LogicalClocks ................................ 10 1.3.6 ClockTicks................................... 12 1.3.7 VirtualClocks ................................ 13 1.4 RepresentationsofTime................................ 14 1.4.1 IntheBeginning .............................. 14 1.4.2 MakingaDateWithPerl....................... 15 1.4.3 High-ResolutionTiming........................ 17 1.4.4 BenchmarkTimers ............................ 18 1.4.5 CrossingTimeZones........................... 19 1.5 TimeDistributions .................................... 21 1.5.1 GammaDistribution........................... 22 1.5.2 ExponentialDistribution....................... 22 1.5.3 PoissonDistribution........................... 24 1.5.4 ServerResponseTimeDistribution .............. 25 xvi Contents 1.5.5 NetworkResponseTimeDistribution ............ 26 1.6 TimingChainsandBottlenecks ......................... 28 1.6.1 BottlenecksandQueues........................ 30 1.6.2 DistributedInstrumentation .................... 30 1.6.3 DiskTimingChains ........................... 31 1.6.4 LifeandTimesofanNFSOperation............. 32 1.7 Failing Big Time ...................................... 33 1.7.1 Hardware Availability.......................... 34 1.7.2 TyrannyoftheNines .......................... 34 1.7.3 Hardware Reliability........................... 35 1.7.4 MeanTimeBetweenFailures ................... 36 1.7.5 DistributedHardware.......................... 38 1.7.6 ComponentsinSeries .......................... 38 1.7.7 ComponentsinParallel ........................ 38 1.7.8 Software Reliability ............................ 39 1.8 MetastableLifetimes................................... 39 1.8.1 Microscopic Metastability ...................... 40 1.8.2 Macroscopic Metastability ...................... 43 1.8.3 Metastability in Networks ...................... 43 1.8.4 QuantumlikePhaseTransitions ................. 45 1.9 Review............................................... 45 Exercises................................................... 46 2 Getting the Jump on Queueing ............................ 47 2.1 Introduction.......................................... 47 2.2 WhatIsaQueue? ..................................... 48 2.3 TheGroceryStore—CheckingItOut .................... 48 2.3.1 QueueingAnalysisView........................ 49 2.3.2 Perceptions and Deceptions . .................... 50 2.3.3 ThePostO?ce—SnailMail .................... 51 2.4 Fundamental Metric Relationships ....................... 51 2.4.1 PerformanceMeasures ......................... 52 2.4.2 ArrivalRate.................................. 53 2.4.3 System Throughput ........................... 55 2.4.4 Nodal Throughput............................. 56 2.4.5 Relative Throughput........................... 56 2.4.6 ServiceTime.................................. 57 2.4.7 ServiceDemand............................... 58 2.4.8 Utilization.................................... 58 2.4.9 ResidenceTime ............................... 59 2.5 Little’sLawMeansaLot............................... 59 2.5.1 ALittleIntuition.............................. 60 2.5.2 AVisualProof................................ 61 2.5.3 Little’sMicroscopicLaw........................ 66 2.5.4 Little’sMacroscopicLaw ....................... 66 Contents xvii 2.6 UnlimitedRequest(Open)Queues....................... 67 2.6.1 SingleServerQueue ........................... 67 2.6.2 MeasuredServiceDemand...................... 68 2.6.3 QueueingDelays .............................. 68 2.6.4 TwinQueueingCenter ......................... 73 2.6.5 ParallelQueues ............................... 74 2.6.6 DualServerQueue—HeuristicAnalysis........... 76 2.7 MultiserverQueue..................................... 79 2.7.1 Erlang’s C Formula............................ 80 2.7.2 AccuracyoftheHeuristicFormula............... 82 2.7.3 Erlang’s B Formula............................ 83 2.7.4 ErlangAlgorithmsinPerl ...................... 84 2.7.5 DualServerQueue—ExactAnalysis ............. 86 2.8 LimitedRequest(Closed)Queues........................ 88 2.8.1 ClosedQueueingCenter........................ 88 2.8.2 InteractiveResponseTimeLaw ................. 89 2.8.3 RepairmanAlgorithminPerl ................... 90 2.8.4 ResponseTimeCharacteristic................... 92 2.8.5 Throughput Characteristic...................... 93 2.8.6 FiniteResponseTimes......................... 94 2.8.7 ApproximatingaClosedQueues................. 95 2.9 ShorthandforQueues.................................. 99 2.9.1 QueueSchematics ............................. 99 2.9.2 KendallNotation..............................100 2.10 ComparativePerformance ..............................101 2.10.1 MultiserverVersusUniserver....................102 2.10.2 MultiqueueVersusMultiserver ..................102 2.10.3 TheEnvelopePlease! ..........................104 2.11 GeneralizedServers....................................105 2.11.1 Infinite Capacity (IS) Server ....................106 2.11.2 Exponential(M)Server ........................107 2.11.3 Deterministic(D)Server .......................108 2.11.4 Uniform(U)Server............................108 2.11.5 Erlang-k (Ek)Server...........................108 2.11.6 Hypoexponential (Hypo–k)Server...............109 2.11.7 Hyperexponential (Hk)Server...................109 2.11.8 Coxian (Cox–k)Server.........................110 2.11.9 General(G)Server ............................111 2.11.10 Pollaczek–Khintchine Formula ..................112 2.11.11 Polling Systems ...............................113 2.12 Review...............................................115 Exercises...................................................116 xviii Contents 3 Queueing Systems for Computer Systems..................119 3.1 Introduction..........................................119 3.2 TypesofCircuits......................................120 3.3 PoissonProperties.....................................122 3.3.1 PoissonMerging...............................122 3.3.2 PoissonBranching.............................123 3.3.3 PoissonPasta.................................123 3.4 Open-CircuitQueues ..................................124 3.4.1 SeriesCircuits ................................125 3.4.2 FeedforwardCircuits...........................125 3.4.3 FeedbackCircuits .............................126 3.4.4 Jackson’sTheorem ............................129 3.4.5 ParallelQueuesinSeries .......................131 3.4.6 MultipleWorkloadsinOpenCircuits ............135 3.5 Closed-CircuitQueues .................................136 3.5.1 ArrivalTheorem ..............................136 3.5.2 IterativeMVAAlgorithm.......................138 3.5.3 ApproximateSolution..........................139 3.6 Visit Ratios and Routing Probabilities ...................140 3.6.1 VisitRatiosandOpenCircuits..................142 3.6.2 VisitRatiosandClosedCircuits.................143 3.7 MultipleWorkloadsinClosedCircuits ...................144 3.7.1 WorkloadClasses..............................144 3.7.2 BaselineAnalysis..............................145 3.7.3 AggregateAnalysis ............................146 3.7.4 ComponentAnalysis...........................150 3.8 WhenIsaQueueingCircuitSolvable?....................151 3.8.1 MVAIsaStyleofThinking.....................152 3.8.2 BCMPRules .................................153 3.8.3 ServiceClasses................................154 3.9 ClassicComputerSystems..............................155 3.9.1 Time-ShareScheduler..........................155 3.9.2 Fair-ShareScheduler...........................157 3.9.3 PriorityScheduling ............................158 3.9.4 ThreadsScheduler.............................160 3.10 WhatQueueingModelsCannotDo......................161 3.11 Review...............................................163 Exercises...................................................164 4 Linux Load Average—Take a Load O?! ....................167 4.1 Introduction..........................................167 4.1.1 LoadAverageReporting .......................168 4.1.2 WhatIsan“Average”Load? ...................169 4.2 ASimpleExperiment ..................................170 4.2.1 ExperimentalResults ..........................172 Contents xix 4.2.2 Submerging Into the Kernel. ....................173 4.3 LoadCalculation......................................174 4.3.1 Fixed-PointArithmetic.........................175 4.3.2 MagicNumbers ...............................176 4.3.3 EmptyRun-Queue ............................178 4.3.4 OccupiedRun-Queue ..........................179 4.3.5 ExponentialDamping..........................180 4.4 Steady-StateAverages .................................183 4.4.1 Time-AveragedQueueLength...................184 4.4.2 LinuxSchedulerModel.........................184 4.5 LoadAveragesandTrendVisualization ..................187 4.5.1 WhatIsWrongwithLoadAverages .............187 4.5.2 NewVisualParadigm..........................187 4.5.3 ApplicationtoWorkloadManagement ...........189 4.6 Review...............................................190 Exercises...................................................190 5 Performance Bounds and Log Jams........................191 5.1 Introduction..........................................191 5.2 Out of Bounds in Florida...............................191 5.2.1 LoadTestResults .............................192 5.2.2 Bottlenecks and Bounds ........................192 5.3 Throughput Bounds ...................................193 5.3.1 Saturation Throughput.........................193 5.3.2 Uncontended Throughput ......................194 5.3.3 OptimalLoad.................................195 5.4 Response Time Bounds ................................196 5.4.1 UncontendedResponseTime....................196 5.4.2 SaturationResponseTime......................196 5.4.3 Worst–CaseResponseBound ...................197 5.5 Meanwhile, Back in Florida ... ..........................198 5.5.1 Balanced Bounds ..............................199 5.5.2 BalancedDemand.............................199 5.5.3 Balanced Throughput ..........................199 5.6 TheX–Files:EncounterswithPerformanceAliens.........201 5.6.1 X-WindowsArchitecture .......................201 5.6.2 ProductionEnvironment .......................202 5.7 CloseEncountersofthePerformanceKind................202 5.7.1 CloseEncountersI:Rumors ....................202 5.7.2 CloseEncountersII:Measurements ..............203 5.7.3 CloseEncountersIII:Analysis ..................203 5.8 PerformanceAliensRevealed............................205 5.8.1 OutofSight,OutofMind......................205 5.8.2 Log–Jammed Performance . .....................207 5.8.3 ToGetaLogYouNeedaTree..................208 xx Contents 5.9 X-Windows Scalability .................................210 5.9.1 MeasuringSiblingX-Events.....................210 5.9.2 SuperlinearResponse ..........................211 5.10 Review...............................................212 Exercises...................................................212 Part II Practice of System Performance Analysis 6 Pretty Damn Quick (PDQ)—A Slow Introduction .........215 6.1 Introduction..........................................215 6.2 HowtoBuildPDQCircuits.............................215 6.3 Inputs and Outputs....................................215 6.3.1 SettingUpPDQ ..............................216 6.3.2 SomeGeneralGuidelines.......................218 6.4 SimpleAnnotatedExample.............................219 6.4.1 CreatingthePDQModel.......................219 6.4.2 ReadingthePDQReport.......................221 6.4.3 ValidatingthePDQModel .....................222 6.5 PerlPDQModule .....................................223 6.5.1 PDQDataTypes..............................223 6.5.2 PDQGlobalVariables .........................224 6.5.3 PDQFunctions ...............................225 6.6 FunctionSynopses.....................................225 6.6.1 PDQ::CreateClosed............................225 6.6.2 PDQ::CreateMultiNode ........................226 6.6.3 PDQ::CreateNode .............................226 6.6.4 PDQ::CreateOpen.............................227 6.6.5 PDQ::CreateSingleNode........................228 6.6.6 PDQ::GetLoadOpt ............................228 6.6.7 PDQ::GetQueueLength.........................229 6.6.8 PDQ::GetResidenceTime.......................229 6.6.9 PDQ::GetResponse ............................230 6.6.10 PDQ::GetThruMax............................231 6.6.11 PDQ::GetThruput. ............................231 6.6.12 PDQ::GetUtilization ...........................232 6.6.13 PDQ::Init ....................................232 6.6.14 PDQ::Report .................................233 6.6.15 PDQ::SetDebug...............................234 6.6.16 PDQ::SetDemand .............................235 6.6.17 PDQ::SetTUnit ...............................236 6.6.18 PDQ::SetVisits................................236 6.6.19 PDQ::SetWUnit...............................237 6.6.20 PDQ::Solve...................................237 6.7 ClassicQueuesinPDQ.................................238 Contents xxi 6.7.1 DelayNodeinPDQ ...........................238 6.7.2 M/M/1inPDQ...............................238 6.7.3 M/M/m inPDQ..............................239 6.7.4 M/M/1//N inPDQ...........................239 6.7.5 M/M/m//N inPDQ ..........................240 6.7.6 FeedforwardCircuitsinPDQ ...................240 6.7.7 FeedbackCircuitsinPDQ......................242 6.7.8 ParallelQueuesinSeries .......................244 6.7.9 MultipleWorkloadsinPDQ ....................246 6.7.10 PriorityQueueinginPDQ......................252 6.7.11 Load-DependentServersinPDQ ................258 6.7.12 Bounds Analysis with PDQ .....................263 6.8 Review...............................................264 Exercises...................................................264 7 Multicomputer Analysis with PDQ ........................267 7.1 Introduction..........................................267 7.2 MultiprocessorArchitectures............................268 7.2.1 SymmetricMultiprocessors .....................269 7.2.2 MultiprocessorCaches .........................270 7.2.3 CacheBashing ................................271 7.3 MultiprocessorModels .................................272 7.3.1 Single-BusModels.............................273 7.3.2 ProcessingPower..............................274 7.3.3 Multiple-BusModels...........................276 7.3.4 CacheProtocols...............................278 7.3.5 IronLawofPerformance .......................287 7.4 MulticomputerModels.................................289 7.4.1 ParallelQueryCluster .........................290 7.4.2 QuerySaturationMethod ......................294 7.5 Review...............................................298 Exercises...................................................299 8 How to Scale an Elephant with PDQ ......................301 8.1 AnElephantStory ....................................301 8.1.1 What Is Scalability? ...........................302 8.1.2 SPECMultiuserBenchmark ....................303 8.1.3 Steady-stateMeasurements .....................305 8.2 PartsoftheElephant..................................306 8.2.1 ServiceDemandPart ..........................307 8.2.2 ThinkTimePart..............................307 8.2.3 UserLoadPart ...............................308 8.3 PDQ Scalability Model. ................................308 8.3.1 Interpretation.................................311 8.3.2 Amdahl’sLaw ................................312 xxii Contents 8.3.3 TheElephant’sDimensions.....................314 8.4 Review...............................................315 Exercises...................................................315 9 Client/Server Analysis with PDQ..........................317 9.1 Introduction..........................................317 9.2 Client/ServerArchitectures.............................318 9.2.1 MultitierEnvironments ........................319 9.2.2 Three–TierOptions............................319 9.3 BenchmarkEnvironment ...............................321 9.3.1 PerformanceScenarios .........................322 9.3.2 WorkloadCharacterization .....................322 9.3.3 DistributedWorkflow..........................324 9.4 Scalability Analysis with PDQ ..........................325 9.4.1 BenchmarkBaseline ...........................326 9.4.2 ClientScaleup ................................333 9.4.3 LoadBalancerBottleneck ......................334 9.4.4 DatabaseServerBottleneck.....................334 9.4.5 ProductionClientLoad ........................335 9.4.6 SaturationClientLoad.........................336 9.4.7 Per-ProcessAnalysis...........................338 9.5 Review...............................................339 Exercises...................................................339 10 Web Application Analysis with PDQ ......................341 10.1 Introduction..........................................341 10.2 HTTPProtocol .......................................341 10.2.1 HTTPPerformance............................346 10.2.2 HTTPAnalysisUsingPDQ ....................347 10.2.3 Fork-on-DemandAnalysis ......................348 10.2.4 PreforkAnalysis ..............................349 10.3 Two-TierPDQModel..................................355 10.3.1 DataandInformationAreNottheSame .........355 10.3.2 HTTPdPerformanceMeasurements .............355 10.3.3 JavaPerformanceMeasurements ................357 10.4 MiddlewareAnalysisUsingPDQ ........................357 10.4.1 ActiveClientThreads..........................359 10.4.2 LoadTestResults .............................359 10.4.3 DerivedServiceDemands.......................360 10.4.4 NaivePDQModel.............................360 10.4.5 AddingHiddenLatenciesinPDQ ...............365 10.4.6 Adding Overdriven Throughput in PDQ..........366 10.5 Review...............................................370 Exercises...................................................370 Contents xxiii Part III Appendices A Glossary of Terms .........................................373 B A Short History of Bu?ers.................................385 C Thanks for No Memories ..................................391 C.1 LifeintheMarkovLane................................391 C.2 ExponentialInvariance.................................392 C.3 ShapePreservation ....................................394 C.4 ACounterexample.....................................394 D Performance Measurements and Tools .....................397 D.1 PerformanceCountersandObjects ......................397 D.2 JavaBytecodeInstrumentation..........................397 D.3 GenericPerformanceTools .............................398 D.4 DisplayingPerformanceMetrics.........................398 D.5 StoringPerformanceMetrics............................401 D.6 PerformancePredictionTools...........................401 D.7 HowAccurateareYourData? ..........................402 D.8 AreYourDataPoissonian? .............................402 D.9 PerformanceMeasurementStandards ....................407 E Compendium of Queueing Equations ......................409 E.1 Fundamental Metrics ..................................409 E.2 QueueingDelays ......................................410 F Installing PDQ and PerlPrograms .........................411 F.1 PerlScripts...........................................411 F.2 PDQScripts..........................................412 F.3 Installing the PDQ Module .............................412 G Units and Abbreviations...................................415 G.1 SIPrefixes............................................415 G.2 TimeSu?xes .........................................415 G.3 CapacitySu?xes......................................415 H Solutions to Selected Exercises ............................417 Bibliography...................................................421 Index ..........................................................427