[rules-users] The "true modify" algorithm - less memory usage, more performance

[Edson Tirelli]

   Hey all

   First of all, I am sorry if I haven't answered many of the e-mails lately
in the list (although I am glad that more and more users are stepping up and
helping others). The reason was we were really busy developing a major
improvement to the Drools algorithm during the last few months and that will
be available in Drools 5.1. For those that haven't seen it yet, we call it
"True Modify" for the lack of a better name:

http://blog.athico.com/2010/03/drools-halves-memory-use-with-new-true.html

   The gains are totally dependent on use cases, but if your application
makes heavy use of update/modify on facts, then you will definitively see
major improvements on memory usage and a nice improvements on performance.
In that particular example we reported in the blog (Drools Planner
Examination Benchmark App) we got ~35% perf improvement and a huge memory
usage improvement: while the original algorithm reached the threshold for
"Full GC" 5 times, the new algorithm never reached the threshold (running
with -Xms256M -Xmx256M).

   The reason for this e-mail is that we have a limited set of test
applications we can run here and I am looking for users that could give the
new algorithm a try to make sure we got everything covered before we release
5.1 final. From a users perspective, it should be 100% backward compatible,
so what I am looking for is for people that can run a benchmark (perf and
memory usage) of your use case using 5.0.1, then using the snapshot from
trunk, make sure the application ran successfully and provide feedback on
the perf/memory usage differences.

   Ping me if you are able to run such a benchmark.

   Thanks,
      Edson

-- 
 Edson Tirelli
 JBoss Drools Core Development
 JBoss by Red Hat @ www.jboss.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/rules-users/attachments/20100330/8d2870a3/attachment.html