Looks like Sun‘s finally posted the SpecJAppserver2004 results they‘ve been working on with us.
This has touched off at least one discussion on TheServerSide, as well as a mention in Jonathan Schwartz‘s blog.
I‘m really excited about these results because:
1) They show decent performance (mind you, this benchmark was run using MySQL-5.0.4 beta, the server is getting faster as we work on performance issues before it ships…seems we have a history of good benchmarks on beta software…Eweek was that way too, with a 4.0 beta…).
The other three node result posted that had slightly better performance, was actually on faster hardware (v20Z‘s with 2.6GHz Opterons, instead of the 2.4Ghz Opterons our test rig had)...If the benchmark would‘ve been on the same hardware, it really looks like we would‘ve had equivalent, if not slightly better performance, considering the fact that the appserver nodes were the bottleneck component on the run that Sun ended up submitting.
2) It shows that MySQL w/ JDBC passes the CTS tests for J2EE (Sun wouldn‘t publish without being able to certify the entire stack), so that‘s a nod to compatibility which is nice to have for those of you that develop portable software out there on top of JDBC. Sun also tells us that this is the first time ever an Open Source database has been used in a CTS-certified stack for this benchmark.
3) It shows that you can get into quite a capable platform without spending a lot of money (I remember back in the .com days, working with folks who were spending as much for one license for an appserver as what this benchmark comes in at for the whole enchilda, including hardware, 24/7 support and licenses).
4) Even though they don‘t post $/JOPS figures for SpecJAppServer2004, ours are really good, considering that you‘re basically paying just slightly south of $70K for hardware, software and 24/7/365 support for 3 years (with MySQL being supported through a MySQL Network Gold subscription)
5) I learned some new stuff…syscalls on Solaris have quite a bit of overhead (but that‘s pretty easy for us to workaround, in fact we now include a configuration “bundle“ with Connector/J 3.1.9 and later that enables the maximum performance mode for Solaris adding “useConfigs=solarisMaxPerformance” as a JDBC URL property does the trick)...DTrace rules (wish we had something like that on Linux), and running the JDBC CTS testsuite on our build on every check-in seems to have been worthwhile ;)
And last, but not least, it was nice to see the results of a lot of hard (but fun) work with the folks over at Sun (Tom Daly and crew), as well as learning more about the internals of InnoDB for me (with Peter Zaitsev, be sure to check out his blog over on LiveJournal if you‘re into MySQL performance stuff).
...Now, if I could just get my wife to let me throw a couple of v20Z‘s in the rack downstairs to play with ;)