brettwooldridge/bitronix-hp

Bitronix-HP (High Performance)

UPDATE: all of the changes in this fork have been incorporated into the official distribution here.

The official Bitronix distribution is located here, so why this distribution?

I am a contributor and committer on the official Bitronix project, created and led by Ludovic Orban. The company I work for and the product we sell has high performance demands, so we are constantly profiling and tuning. We have been using Bitronix for more than 5 years, and during profiling runs, I got so used to Bitronix showing up in the list of "hotspots" that I put a filter into my profiler to basically ignore all Bitronix classes so I could focus on improving the performance of our application code.

However, as the application got faster and faster (due to profiling and tuning), just this past year I turned off the filter for Bitronix and to my surprise, I found Bitronix was now becoming a bottleneck. I had contributed some significate pieces of code to Bitronix in the past, so being familiar with the code, I dug in. Bitronix has quite a lot of high-level locking using synchronized constructs, which impedes performance. The decision of the Bitronix project to move off of JDK 1.4 and to JDK 1.5 allowed me to take advantage of some of concurrency improvements there, and in fact my goal became to remove every 'synchronized' keyword I could find, while still maintaining correctness.

The result is Bitronix-HP. So, still the question: why this distribution? I like Ludovic, and I like working on Bitronix, but sometimes developers disagree. My changes touched a lot of code. I almost completely rewrote the XAPool class. I removed the use of reflection that was used as a shim to support JDBC3 and JDBC4 targets while still compiling against only JDBC3 interfaces. I wrote that hairy code originally, so it was a pleasure to rip it out.

I can't say exactly why Ludovic has not taken these improvements, but I think it is a combination of things. First, he has stated he doesn't have the bandwidth to review all of these changes. Second, because of that he is reluctant to just merge them given that Bitronix is 'stable' and doesn't want to destabilze the code. And finally, some of the changes (compiling JDBC4 stubs) increased the complexity of the build, which he doesn't like. But let me be clear, I would like nothing more than to see all of the improvements, including the new JDBC4 support (a big performance win), incorporated back into the Bitronix mainline.

I put this repository here, not just for other people, but also because I (and my company) don't like sitting on private changes to an open source project. If I get hit by a bus, finding the improvements we made to Bitronix should be easy.

Is it stable?

I would say, absolutely. I say that because my company has our product deployed in some of the largest telecom providers in the world running high transaction rates, and running (24/7) on this code-base for the past year. All test cases in the Bitronix suite pass, of course.

Technical Details

You can find a technical analysis of Bitronix-HP (aka BTM-2.2 branch) and Bitronix 2.1.2 here. Included in the analysis is a detailed description of the XAPool changes.

Do you need it?

Whether you need bitronix-hp will depend largely on your workload. Only you can determine whether it improves performance in your environment. I would say test, and then test again. Measure CPU, overall time for important workloads, etc.

Installation

Bitronix-HP is intended to be 100% compatible with Bitronix. We regularly merge changes from the Bitronix mainline.

Install Bitronix first, and work out all of the kinks in your configuration -- Spring, JPA/Hibernate, whatever. Only after everything is working, trying dropping in Bitronix-HP and then test your performance.

In the download area you will find a jar:

btm-2.2.0-SNAPSHOT.jar

Replace your Bitronix 2.1.x jar file with the btm-2.2.0-SNAPSHOT.jar. Bitronix-HP is not available in a maven repository at this time.

If you ever encounter an issue with the Bitronix-HP fork, I recommend dropping back to the standard Bitronix distribution to see if the problem also occurs there.

Forking

Basically, I would prefer if you did not fork this project. Why? Because I think all meaningful and significant changes should be contributed to the core Bitronix project. Only if there is an improvement that is specific to the Bitronix-HP code would I recommend that you fork this project. Of course, if you do fork it, I welcome Pull Requests.