Analyzing Computer System Performance with Perl::PDQ (eBook)

  Neil Gunther, M.Sc., Ph.D. is a leading industry computer performance consultant who founded Performance Dynamics Company (www.perfdynamics.com) in 1994. Prior to that, Dr. Gunther held research and management positions at San Jose State University, JPL/NASA (Voyager and Galileo spacecraft), Xerox PARC research center and Pyramid/Siemens Technology. His performance and capacity planning classes have been presented at such organizations as Boeing, Fedex, Motorola, Stanford University, and Sun Microsystems. Dr. Gunther is the author of over a hundred papers on computer performance topics, including "HIT-AND-RUN TACTICS ENABLE GUERRILLA CAPACITY PLANNING," published recently in IEEE IT Professional magazine. He has also written the book THE PRACTICAL PERFORMANCE ANALYST (iUniverse.com 2000). In 1996 he was awarded Best Technical Paper at CMG, and in 1997 was nominated for the A.A. Michelson Award. Dr. Gunther is a member of the AMS, ACM, CMG, IEEE, SAGE-AU, SIGMETRICS and USENIX.

Contents 13
Preface 6
Part I Theory of System Performance Analysis 22
1 Time—The Zeroth Performance Metric 23
1.1 Introduction 23
1.2 What Is Time? 24
1.3 What Is a Clock? 28
1.4 Representations of Time 34
1.5 Time Distributions 41
1.6 Timing Chains and Bottlenecks 48
1.7 Failing Big Time 53
1.8 Metastable Lifetimes 59
1.9 Review 65
2 Getting the Jump on Queueing 67
2.1 Introduction 67
2.2 What Is a Queue? 68
2.3 The Grocery Store—Checking It Out 68
2.4 Fundamental Metric Relationships 71
2.5 Little’s Law Means a Lot 79
2.6 Unlimited Request (Open) Queues 87
2.7 Multiserver Queue 99
2.8 Limited Request (Closed) Queues 108
2.9 Shorthand for Queues 119
2.10 Comparative Performance 121
2.11 Generalized Servers 125
2.12 Review 135
3 Queueing Systems for Computer Systems 138
3.1 Introduction 138
3.2 Types of Circuits 139
3.3 Poisson Properties 141
3.4 Open-Circuit Queues 143
3.5 Closed-Circuit Queues 155
3.6 Visit Ratios and Routing Probabilities 159
3.7 Multiple Workloads in Closed Circuits 163
3.8 When Is a Queueing Circuit Solvable? 170
3.9 Classic Computer Systems 174
3.10 What Queueing Models Cannot Do 180
3.11 Review 182
4 Linux Load Average—Take a Load Off! 185
4.1 Introduction 185
4.2 A Simple Experiment 188
4.3 Load Calculation 192
4.4 Steady-State Averages 201
4.5 Load Averages and Trend Visualization 205
4.6 Review 208
5 Performance Bounds and Log Jams 209
5.1 Introduction 209
5.2 Out of Bounds in Florida 209
5.3 Throughput Bounds 211
5.4 Response Time Bounds 214
5.5 Meanwhile, Back in Florida . . . 216
5.6 The X–Files: Encounters with Performance Aliens 219
5.7 Close Encounters of the Performance Kind 220
5.8 Performance Aliens Revealed 223
5.9 X-Windows Scalability 228
5.10 Review 230
Part II Practice of System Performance Analysis 231
6 Pretty Damn Quick (PDQ)—A Slow Introduction 232
6.1 Introduction 232
6.2 How to Build PDQ Circuits 232
6.3 Inputs and Outputs 232
6.4 Simple Annotated Example 236
6.5 Perl PDQ Module 240
6.6 Function Synopses 242
6.7 Classic Queues in PDQ 255
6.8 Review 281
7 Multicomputer Analysis with PDQ 283
7.1 Introduction 283
7.2 Multiprocessor Architectures 284
7.3 Multiprocessor Models 288
7.4 Multicomputer Models 305
7.5 Review 314
8 How to Scale an Elephant with PDQ 316
8.1 An Elephant Story 316
8.2 Parts of the Elephant 321
8.3 PDQ Scalability Model 323
8.4 Review 330
9 Client/Server Analysis with PDQ 332
9.1 Introduction 332
9.2 Client/Server Architectures 333
9.3 Benchmark Environment 336
9.4 Scalability Analysis with PDQ 340
9.5 Review 354
10 Web Application Analysis with PDQ 355
10.1 Introduction 355
10.2 HTTP Protocol 355
10.3 Two-Tier PDQ Model 369
10.4 Middleware Analysis Using PDQ 371
10.5 Review 384
Part III Appendices 385
A Glossary of Terms 386
B A Short History of Buffers 397
C Thanks for No Memories 402
C.1 Life in the Markov Lane 402
C.2 Exponential Invariance 403
C.3 Shape Preservation 405
C.4 A Counterexample 405
D Performance Measurements and Tools 407
D.1 Performance Counters and Objects 407
D.2 Java Bytecode Instrumentation 407
D.3 Generic Performance Tools 408
D.4 Displaying Performance Metrics 408
D.5 Storing Performance Metrics 411
D.6 Performance Prediction Tools 411
D.7 How Accurate are Your Data? 412
D.8 Are Your Data Poissonian? 412
E Compendium of Queueing Equations 418
E.1 Fundamental Metrics 418
E.2 Queueing Delays 419
F Installing PDQ and PerlPrograms 420
F.1 Perl Scripts 420
F.2 PDQ Scripts 421
F.3 Installing the PDQ Module 421
G Units and Abbreviations 423
G.1 SI Prefixes 423
G.2 Time Suffixes 423
G.3 Capacity Suffixes 423
H Solutions to Selected Exercises 425
Bibliography 429
Index 435