JRuby 1.7.0dev Performance Test

Following Charles call for tests on jruby 1.7 dev,
I’ve tested one of my internal systems through a variety of ruby
interpreters and several java / jruby flags.

Where are the results

Though I don’t have permission to release the code, I’m working on
reducing the test case to something
business agnostic which I will be able to share.

If any of you guys have any pearls of wisdom regarding which flags will
make jruby faster, I would be glad
to given them a try and report back the results.

Cheers,
Lus Landeiro R.

Thank you, this is interesting!

Another flag to throw into the mix would be -X+C, which forces each
.rb file to compile to JVM bytecode immediately, rather than
interpreting for a while. This could improve perf in some cases.

AR is an interesting target for benchmarking. By interesting, I mean
frustrating. For years JRuby has been faster than the C impls, but AR
on JRuby remained considerably slower. Over time we’ve improved that,
but it has never been completely clear what we were fixing along the
way. The truth may be that AR is so object/memory-intensive that our
incremental improvements to JRuby and AR-JDBC have slowly but surely
bumped performance up.

I’m interested in the invokedynamic perf here, since it shows there’s
bottlenecks that either we or the JVM guys need to fix. Thank you for
posting this!

  • Charlie

2011/8/5 Luís Landeiro R. [email protected]: