Customer Reviews

Performance Tuning for Linux(R) Servers

5 star:		(1)
4 star:		(2)
3 star:		(2)
2 star:		(2)
1 star:		(0)

 

 

IBM continues to unify its operating systems under linux, and to promote its linux-based expertise to others. This book is a good demonstration of its commitment. You are shown how to do detailed benchmarking. Which is necessary before you can do any serious tuning. For if you can't measure some property, improvement of it is harder.

The latest linux kernel, 2.6, is covered. Though much of the advice also pertains to earlier versions, especially 2.4.

En route, the book discreetly plugs IBM's java implementation and JDK. The book is impressive in its restrained treatment [ie. touting] of IBM's products. It's certainly not a sales brochure. The advice is broadly applicable across any hardware that runs linux.

I like this book since it does give some really good hints at performance tuning Linux boxes. I will say if all your systems are working right is time to buy this book. If it is not do not buy this book

If you're responsible for Linux servers and you need to keep things running at peak efficiency, you *need* to get this book... Performance Tuning For Linux Servers, edited by Sandra K. Johnson, Ph. D., Gerrit Huizenga, and Badari Pulavarty. It's an excellent blend of theory and practicality.

Chapter List:
Part 1 - Linux Overview: Linux Installation Issues; Kernel Overview; Overview Of Server Architectures
Part 2 - Performance Analysis Tools: System Performance Monitoring; System Trace Tools; Benchmarks As An Aid To Understanding Workload Performance
Part 3 - System Tuning: System Performance Principles And Strategy - A Benchmarking Methodology Case Study; Scheduler Tuning; The Linux Virtual Memory-Performance Implications; I/O Subsystems-Performance Implications; File System Tuning; Network Tuning; Interprocess Communications; Code Tuning
Part 4 - Performance Characterization Of Linux Server Applications: Web Servers; File And Print Servers; Database Servers; Application Servers
Part 5 - Tuning Case Studies: Case Study - Tuning The I/O Schedulers In Linux 2.6; Case Study - File System Tuning; Case Study - Network Performance On Linux; Case Study - Commercial Workload Tuning; Tuning Kernel Parameters; Index

This is one of those rare books that ranks high on many criteria... It's got a lot of theory, the "why" of different features as they relate to performance. It's also packed full of practical material. They tell you how to measure key components in the system and what parameters you can change to affect those areas. When you get done, you've covered every conceivable area that exists in the Linux environment. I particularly liked the blending of measuring/monitoring along with the instructions on how to change performance. The chapter on system performance monitoring tools can be used immediately to see how your system runs. Once you become comfortable using those tools, you can use them to run before and after comparisons of tuning efforts. There's no guesswork involved. Measure, tweak, and remeasure. Repeat as necessary. Same with the benchmarking tools. They will allow you to know without a doubt whether the changes you made work or not.

I'm also impressed with the readability of the book on a couple of different fronts. For one, books like this can be dry as dirt, especially in the areas dealing with theory and architecture. But surprisingly, it really wasn't hard to follow, nor was I getting bored. The other reason the readability is surprising is that the chapters are done by 21 different contributors. When you get different chapters being done by different techies (with varying levels of communication skills), you normally get a very uneven book. Not so here... The editors should be commended for taking material that I'm sure was all over the board and weaving it into a cohesive and coherent volume.

Bottom line... If you run a Linux system and you are responsible for making sure it performs, this is the book you need to have.

I just can't bring myself to like this. I did find it uneven in spite of the editors obvious attempts at blending. I want to like it - individually the chapters are excellent, but I just have trouble seeing this as a coherent work. For me, it's choppy and the different styles make me uncomfortable.

Perhaps I can see it as a reference book. I could see myself pulling this down and going to a particular chapter for advice on that specific subject. As such, this surely has value, but I just don't care for it as a corpus.

I bought this book because of some of the names involved in this book (both authors and contributors).

However I share the similar thoughts with some other reviewers after spending significant amount of time reading the book. Every chapter seems to start from the beginning. Every authors (of each chapters) try to explain things from scratch. Some jump into deaitls with unexplained term very quickly, and if you are not already familar with kernel sources codes, or have done other researches, you will be lost. Some barely touch the surface, and again if you don't have prior experience, you will be confused and maybe misinformed.

The book is suppose about Performance. But there are a lot of tunings and advice about sizings. If a webserver tells a user it is busy and cannot serve any page, that's a sizing problem, not a performance problem. If a server drop new network request because it has run out of resource (not because it's slow and timed out), it's not a performance problem either.

Performance is also not about benchmarking. I don't see the use of large amount of numbers in the case studies. They may be significantly different when performance using different kernel version, and different hardware. This should not be a technical paper where numbers are important. The Authors should explain why or how something can be done to improve the "performance" and the side effects.

And there are simple mistakes, and typo. Seems like someone collected all these notes and try to edit them. But obviously no one person or a small team of person is good enough to find those mistakes.

Each chapter is written by a different person, and as result there's a rather disparate feel to the whole book. It's more like collection of scientific paper, authored by stuffy prefessors cooped up in a dusty office somewhere. The words are interesting and well written, and there's plenty to learn, but if you're expecting a group of quick hacks to impress your boss on Monday morning, you'll be disappointed. You will however, be a better and wiser systems administrator.

for more book reviews, read <a href="http://www.street71.com/techbook/2005/12/26/performance-tuning-for-linux-servers/">Tech Books Reviews</A> on Performance Tuning for Linux Servers.

I found this book to be uneven, and to have problems getting the details right, even in some simple cases. I really wanted to like this book, and I hope IBM can put together a second edition that fixes the serious flaws this one has.