ObjectWeb Consortium
Search ObjectWeb Mail Archive: 

Advanced Search - Powered by Google

Mail Archive Home | proactive List | August 2007 Index

<--  Date Index  --> <--  Thread Index  -->

Re: [proactive] JVM Benchmark results

Some comments:
 * Did you measure the time spent on GC for these runs?

No. But I just re-ran the tests on one machine (mine: petawawa) with -XX:+PringGCDetails and GC takes <15ms across all the Sun JVMs, so I'd say it is not a consequential part of the compute time.

* Are all these VMs starting with the same default configuration (min/max heap size....)

I used the defaults, however during the re-run on petawawa with fixed -Xms128M -Xmx512M for all JVMs I got equivalent results.

* Since your code seems to rely on random number, are you sure these VMs use the same systems calls to generate them? On linux, using /dev/random on a machine with no user in front can cause unexpected pauses in the execution

Yes, we are using a pure-java implementation of a PRNG called the Mersenne Twister (Fast). In some benchmarks, it is twice as fast as the java.util.Random. I did a little reading about java.security.SecureRandom, but I don't see how to select /dev/urandom, and I wouldn't know how to "test" if the quality of the (P)RNG was suitable for our application -- we know that the Mersenne Twister algorithm is good, so we're using it.



Ian.Stokes-Rees@xxxxxxxxxxxxxxx                 T: +33 4 92 38 71 64
INRIA Sophia, France                            F: +33 4 92 38 76 44
Project OASIS/ProActive - Grid Research Team    http://inria.fr/oasis
org:INRIA Sophia, France;Project OASIS/ProActive
adr:;;2004 Rte Des Lucioles BP 93;Sophia Antipolis;;06902;France
title:Research Scientist
tel;work:+33 4 92 38 71 64
tel;fax:+33 4 92 38 76 44

<--  Date Index  --> <--  Thread Index  -->

Reply via email to:

Powered by MHonArc.

Copyright © 1999-2005, ObjectWeb Consortium | contact | webmaster.