Also sequential is aimed for stateless services, it's rule building time
is a slighter longer. So Once the RuleBase is built try running the test
a few hundred times, once Sequential and once Rete. Also check memory,
it should use less memory. Its hard to get any real perf numbers in
3.44s. All tests need to run under load for preferably 2 minutes, this
allows for several generations of GC to kick in, which is what normally
hampers performance in a Rule System. If sequential uses less memory and
makes less objects its real advantage will kick in once GC happens.
To help with normalisation I often don't do how long for X iterations,
instead i say how many iterations in 2 minutes.
Mark
Mark Proctor wrote:
If the rules and data don't push the system then you won't see a huge
amount. As the rules and data grow in size, but we are talking 50K
facts and thousands of rules to see a 350% increase. I'm surprised a
RuleBase with a 1000 rules only sees a 2% increase though, while not
large its still at the lower end of medium. If the data is small and
the rules well written then that could account for it.
but the following, which you are doing, is correct:
RuleBaseConfiguration conf = new RuleBaseConfiguration();
conf.setSequential(true);
RuleBase rb = RuleBaseFactory.newRuleBase(conf);
StatelessSession session = rb.newStatelessSession();
Are you doing any function calls or reasoning that relies on outside
calculations? That could account for it, if the bottleneck is outside
of the engine reasoning, the advnatages gained will not be noticeable.
Mark
Jin, Ming wrote:
My rules engine test showed virtually no difference between Rete and
Sequential execution modes: 3.44s for Rete while 3.50s for Sequential.
Those are the average timing from several runs, each with 1,000 rule
execution calls. Since I do not have a controlled environment, the
difference seems insignificant. (The other rules engine I tested showed
huge improvement for the same rule set comparing Sequential with Rete).
The following code segments show how RuleBaseConfiguration is set and
the rule execution. Traced into the code a little bit, and did see the
sequential flag set and the order() method was called.
Please advise if this is the proper way to set sequential. Otherwise, I
would just conclude that sequential mode has no effect for this
particular rule set under Drools.
I would like to add that even without sequential, Drools performance was
fantastic.
Thanks,
-Ming
/**
* Creates a RuleBase instance with the given package object and the
sequential execution flag.
* @param pkg
* @param sequential true for sequential mode, false defaults to Rete
* @return the created RuleBase with the given package and execution
mode. * @throws Exception
*/
private static RuleBase createRuleBase(Package pkg, boolean
sequential) throws Exception {
RuleBaseConfiguration conf = new RuleBaseConfiguration();
conf.setSequential(sequential);
RuleBase rb = RuleBaseFactory.newRuleBase(conf);
rb.addPackage(pkg);
return rb;
}
// code segments for rule execution
StatelessSession session = rulebase.newStatelessSession();
StatelessSessionResult result =
session.executeWithResults(list);
_______________________________________________
rules-users mailing list
[email protected]
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
[email protected]
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
[email protected]
https://lists.jboss.org/mailman/listinfo/rules-users
