[concurrency-interest] Some interesting (confusing?) benchmark results

Aleksey Shipilev aleksey.shipilev at oracle.com
Mon Feb 11 16:19:48 EST 2013


On 02/12/2013 12:52 AM, thurston at nomagicsoftware.com wrote:
> Does it mean I'm doing something wrong?

Probably. Tuning FJP to fit the needs of the problem at hand is
something tricky to do for unexperienced user.

> I'll give a brief description of my implementation:

Complete source for the test will be helpful.

Also, try to use this:
  https://github.com/shipilev/java-forkjoin-trace
...to analyze the behavior of your workload.

> Just ran it on my laptop:
> Windows 7 64-bit
> 1.7.0_03; Java HotSpot(TM) 64-Bit Server VM 22.1-b02
> Core 2 Duo==> 2 cores 2GHz

I'm not mocking your machine, but your JDK version. That version is
terribly outdated, and probably contains quite old FJP code. Pull the
fresh JAR from here:
 http://g.oswego.edu/dl/concurrency-interest/
or here:
 http://shipilev.net/builds/jsr166/

...then -Xbootclasspath/p:<path> it. (check with -verbose:class you have
indeed loaded FJP from that JAR). Although you might have problems
running in on non-JDK8-lambda build. In that case, you can fallback to
java-forkjoin-trace JAR, which has recent enough FJP (but not the
bleeding edge to be incompatible with JDK7). Doug can have more ideas
how to triage FJP perf issues.

-Aleksey.


More information about the Concurrency-interest mailing list