The Art of Computer Systems Performance Analysis: Techniques for Experimental Design, Measurement, Simulation, and Modeling 1st Edition

I had to buy this book as main recommended reference for one of the compulsory subjects of my graduate studies. My focus and interests areas are more into Computational Science and Machine Learning so I honestly wasn't that eager to study it. The coverage on performance analysis is excellent. I learned a lot from it e.g. Queuing Theory. I recommend it for this purpose, however this is not the focus of my review.

I own most of the best books in the areas of Machine Learning and Statistics. It really amaze me how this book alone with a very uncluttered and pragmatic way clearly explains and support with detailed step by step examples what most of the other best books in those areas miserably fail to show. This book offers one of the best introductions to Statistics I know of e.g. explanation of t-test, chi-square test, confidence intervals, ANOVA etc. There is really no better book I know for explaining what PCA is all about ... all the Machine Learning books I own spend many pages even chapters and fail to clearly show the concept this book do show in just a couple of pages ... really impressive!

Don't be fooled by the publication date, the concepts are still very relevant and there is no book on Statistics I can recommend better than this one. Plus you will learn statistics with excellent performance analysis examples. This is the perfect mix to have e.g. software developer taking a refresher in Statistics. However, do pay attention to the *reprint* date as there are multiple prints around and the errata is quite large e.g. avoid buying an old print from the "Marketplace".

This is the required text in a CS graduate class on Computer Performance Modeling that I have taken couple of months ago. For the most part, the concepts are presented in fairly clear fashion and are pretty easy to understand. As another reader noted, the math in this book is not that hard. We went through about 90% of the book in class. One downside of the book is that there a bunch of errors for which errata is available on Raj Jain's site. Our Professor teaching the class mentioned that he talked to Raj Jain about fixing the errors in a reprint but apparently he was busy with other projects, so this didn't happen. Since the errata is fairly big, I would suggest you updated the relevant areas of the text before starting to read. In the class, we also used R package for some of the statistical parts (ANOVA for example). While this book is pricey, I am sure it will be useful beyond the class in the industry, so I am keeping it. I certainly agree with the other reader who said that every CS/EE person should read this work. Highly recommended.

I used Jain's performance book for my performance modeling course. It is very thorough in terms of the topics it covers, and I like that. Sometimes it is rather straight forward, and other times it is difficult to follow, and I think that could be improved upon. The book has content that is very practical as well as more academic content, which I liked.

There are far too many errors in the book, and that is quite frustrating. There is plenty of errata, but a new edition of the book should be made with corrections. The book is rather dated, and I wish that in the new edition, it would be placed in the most modern context possible. I would like a companion site to go with the book where readers could discuss difficult material.

This old book is one of the most valued on my shelf. I was first exposed to it in a graduate class, and I have to say that the book is so good I was not aware that my professor was not a good instructor until I had him a second time in a class where the textbook was less than stellar. Don't judge the book by a quick perusal either. At first glance, especially if you are looking at Part I, it looks like one of those books on Six Sigma that will put you to sleep. In fact, the vast majority of the book is quite interesting.

Part II, "Measurement Techniques and Tools", are where things get interesting. The good part about this entire book is that it uses problems in the analysis of computer systems as the basis of presentation for all tools presented. The graphs are excellent, the mathematics are largely self-contained, and if algorithms are presented they are usually given in numbered steps and an actual computer program shown. This is one drawback of the book - it uses the ancient Simula language for its demonstration code. However, if you are familiar with C, Java, or any of the other mainstream procedural languages, you'll find that Simula looks like very readable pseudocode, so this should not be an obstacle to understanding.

Part III is a section dedicated entirely to probability theory and statistics. Starting with the simple definition of the mean, this handy section not only derives all of the statistics you need in this book, it talks about common mistakes made in applying them.

Part IV is about experimental design and analysis. Using the mathematics developed in part three this section talks about all aspects of designing a proper experiment for the measurement or simulation of a computer system, including common mistakes and the best choice for the size of your experiment.

Part V presents the key issues in simulation modeling. First it discusses simulation terminology, simulation design criteria, and stopping conditions. Random number generation is the subject of three chapters in reference to inputs to your simulation. Finally there is a chapter on the commonly used distributions such as Bernoulli, beta, binomial, etc. that talks specifically about random number generation algorithms for each of the distributions presented. What makes this section so valuable is that although you may have possibly seen the math before, more than likely you don't know the value of each kind of distribution. This section makes that issue clear in terms of modeling computer performance.

Part VI is on queuing models, and is probably the most difficult section in the book. Although it is one of the better written pieces I have read on queueing theory, it is not as easily grasped as previous sections based on reading the textbook alone. There are examples present, and the book does a good job of presenting "the big picture" as to the use of queueing theory in computer performance analysis, but you may need outside material to really grasp how to set up a queueing problem from a mathematical standpoint.

No other book I've found does such a good job of discussing all of the topics covered and clearly tying it into practical issues in measuring and monitoring system performance. I highly recommend it.