Back to newsletter 047 contents

Well, 5.0 was released. Personally I'm entirely unconcerned by Sun calling this the "5.something" release. My concern is never with the name, it is with the performance and robustness of the release. And unfortunately, as usual, early indications are that the ".0" release is only stable enough for testing, not deployment.

That's not Sun's fault, it is endemic in our industry. When was the last time the ".0" release of any software product was so stable that it wasn't quickly followed by the ".1" or ".0.1" or some such patch release? I'm sure there are some, but they are few and far between. It's because there are no penalties to releasing buggy software. Hardware engineers releasing faulty hardware have to recall the product and re-issue the new version, normally at significant expense. Software engineers can usually just re-issue a patch, or new version at very little cost - to the vendor that is.

Is this a problem? Not really for the ".0" releases, everyone knows these are really beta releases. But the oft quoted figure that maintenance costs four times as much as development gives an indication for what the real cost of software would be if we factored in sufficient robustness and performance testing. I don't know whether hiding the real cost of software in this way is a problem. I do suspect that if everyone priced their software to produce fully robust and efficient ".0" versions, our industry would be much smaller and more specialized than it is.

A note from this newsletter's sponsor

In the newsletter we list our usual raft of articles, news, and we provide all our usual sections. Kirk covers 5.0 performance, a surprising memory difference, J2EE monitoring, and much more in his roundup; Kirk also awards another Meadow Muffin award for practices that should not be repeated. Javva The Hutt introduces us to "stupidity" bottlenecks; and our Question of the month compares EJB performance vs. JDBC direct calls.

This month we focused on some java.util.concurrent related articles, and as a result we have many new java.util.concurrent performance tips extracted in concise form. Also, I forgot to list last month's article on The Performance Interview here, so that's the link for those of you who missed it.

A note from this newsletter's sponsor

News

Java performance tuning related news.

• 5.0 has been tested with heaps of over 500 GBytes & 100 processor machines. Also includes performance ergonomics, which analyze the system at startup and auto configures the JVM for the most appropriate settings.
• 5.0 has improved startup time; server side performance greatly improved on machines with two or more CPUs; bytecode insertion added for profiling
• Sun, Arm partner to improve Jazelle performance
• java.util.concurrent backport for 1.4
• Yet another TMC .NET vs J2EE study, controversial as usual. You might wonder why they bother - it's because they are well paid by Microsoft to do it.
• In the real world J2EE scales better when complexity is involved.
• Line reading speed in Java/Python/C. Annoying that so many people had a go and almost no one could write decent optimizations. Have a look at my book.
• Jetty 5 announcement and consideration of architectural performance improvements
• The resultant discussion considering Jetty 5 speed and alternatives

A note from this newsletter's sponsor

Tools

• Weblogic Portal Cache Monitoring Tool
• Method Scheduler for Java, execute methods asynchronously and/or repeatedly.
• VolatileHashMap (WeakHashMap with weak values rather than keys), int key HashMaps, 5.0 java.util.concurrent backports, connection pool, socket wrappers, in-process sockets, various Stream classes (Fork, redirecting, tee, null, compressed)

Recent Articles

• The ABCs of Synchronization, Part 1 (Jeff Friesen/java.net)
• The ABCs of Synchronization, Part 2 (Jeff Friesen/java.net)
• Java Threads, 3rd Edition, Chapter 5 (Scott Oaks, Henry Wong/O'Reilly)
• Java Threads, 3rd Edition, Chapter 6 extract (Scott Oaks, Henry Wong/O'Reilly)
• Improve J2EE Application Performance with Caching (Scott Robinson /developer.com)
• J2EE Design Patterns (Alan Baumgarten/BEA)

Jack Shirazi

Back to newsletter 047 contents