[concurrency-interest] Intel's HLE and Java

Doug Lea dl at cs.oswego.edu
Thu Apr 26 07:44:09 EDT 2012


On 04/26/12 07:10, Andrew Haley wrote:
>  One thing that I mentioned in my previous mail is that
> there is a pressing need for some way of measuring all this with a
> reasonably representative set of benchmarks.  Did you have any such
> thing when developing j.u.c ?
>

Lots of microbenchmarks (some of which are available in our
src/test/loops CVS), that try to reduce expected use cases
to simply measurable tests. Sometimes they are obvious --
For example, anything with messages/queues can be tested with
programs that omit everything going on in a program except the
producer/consumer loops. Sometimes not -- for example, there are
too many different use patterns of concurrent maps to capture
with confidence in only one or a few tests.

There's still a need for benchmarks that better capture
application-level and integration issues for parallel workloads.
The new/upcoming specJBB2012 finally (unlike its predecessors)
looks like it will be a pretty good tool for evaluating
possible library and VM improvements.

-Doug


More information about the Concurrency-interest mailing list