Java Platform Performance: Strategies and Tactics [Paperback]

The advice is generally sound, such as "use a profiler to decide what to optimize", but all too often the advice stops there. One learns that linked lists are fast for deletes and array lists are fast for random access. But what if neither has the performance characteristics you need?

I would recommend Jon Bentley's "Programming Pearls" for someone who really wants to get a feel for optimization through a number of tight case studies. The examples may be C, but they're insightful and more applicable to the problems you're likely to face than the ones encountered when writing the Swing or Collections classes themselves. Then go out and use a serious optimizer like Quantify, which is essentially what the authors of this book suggest. At that point, you'll wonder why the authors stopped their review of Collections with Sun's built-in implementations.

The code for running mulitple programs and controlling class loading in Chapter 6, though I have not tried it, is interesting and something I'd not seen elsewhere. Other diversions, such as timers and threading in Swing are covered elsewhere. What would have been helpful is a detailed look at debugging threading performance, say through object, resource or thread pooling (an idea dismissed earlier for "small objects").

One glaring omission is a serious micro-benchmarking of basic operations. I found the comparison of speed when using final vs. non-final classes or methods to be startling in practice, and numbers here would help. Similarly, access times for hash tables vs. arrays, simple object construction times, floating point vs. integer arithmetic times, object variable vs. local variable access, etc. There are some wonderful micro-benchmark applets on the web with this functionality. What you'll learn is that up-casting takes forever, so any use of built-in collections/iterators for performance-critical operations will do you in.

I found the constant sales pitch for the Sun HotSpot (TM) Virtual Machine a bit tiring. It concludes with a recapitulation of the marketing materials in Appendix B. It is alternately described as "highly optimized", "state-of-the-art", having a "superior memory allocator", "ultra-fast", having "excellent multiprocessor performance characteristics", having "agressive inlining", having "excellent paging", etc. etc. The appendix devoted to HotSpot (TM) is particularly noteworty for its omission of any profiling information!

The first part on strategies is fundamental and cover concepts on performance. This is a must for anybody who wants to get started on the topic.

The second part on tactics goes into details about spesific issues in Java, and because of the GUI / Swing background of the authors, it is mostly focused on topics related to that. If that's what you're working on, this is a 5 star book, but since Java is becoming more and more server focused technology, it's a little bit missing for people like myself who write Servlets and JSPs.

Hopefully, there'll soon be a book that discusses performance in J2EE.

Wilson and Kesselman have done an excellent job of getting under the bonnet (or hood for those in the US) of Java whilst firmly remaining practical rather than theoretical.

I bought the book because we suspected memory leakage problems in one of our applets but couldn't find much information on the subject elsewhere. The net result is that we've had our eyes opened to a number of issues we weren't aware of and have been able to tackle them with informed confidence rather than ill-educated guesswork.

A particular bonus is the information on benchmarking. So often we find we develop solutions on the basis of ease of coding rather than performance (not always bad thing). What I've found now is that I have greater confidence in both trying out different solutions and then concrete evidence with which I can justify my decisions to others.

Though not for beginners, I would reccommend this book as a good one for expanding your knowledge of how to deliver good, practical, faster and more robust Java. Being able to give sound background information on why you write a piece of code in a particular way is an important step on the way from being a developer to being a key developer.