3:52 p.m.
I am trying to implement a performance monitor to find the most problematic
rules in order to optimize performance. The main performance metrics I want
to gather are:
1. How any times a rule's conditions have been evaluated
2. How long each rule's condition evaluation took
3. How many times a rule's then clause actually executed
4. How long execution of a rule's then clause took
Are there any existing or recommended approaches to do this? I have looked
some at the AgendaEventListener methods but can't clearly map events to the
above. Is there some other logging interface with more events?
Thanks in advance.
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Performance-monitorin...
Sent from the Drools - User mailing list archive at Nabble.com.
6:43 p.m.
On 05/11/2010 20:52, H.C. wrote:
I am trying to implement a performance monitor to find the most
problematic
rules in order to optimize performance. The main performance metrics I want
to gather are:
1. How any times a rule's conditions have been evaluated
2. How long each rule's condition evaluation took
1 and 2 are hard because
it's not that straight forward, it's not a
single block. And the tree traversal approach to our current Rete impl
means it interates from root to tip repeatedly, so you can't easily
determine how long it spends in each node, you could probably add a
counter but this would be custom code. We ar working on an new algorithm
that is single pass, more like a pipeline, which will make ti easier to
get statistics and timings.
3. How many times a rule's then clause actually executed
4. How long execution of a rule's then clause took
That you can do with the
agenda listener. Also try looking at the audit
log impl.
Are there any existing or recommended approaches to do this? I have
looked
some at the AgendaEventListener methods but can't clearly map events to the
above. Is there some other logging interface with more events?
Thanks in advance.
3:59 p.m.
Mark,
Thanks for your reply. I figured 3 and 4 were currently possible but 1 and 2
puzzled me.
"it interates from root to tip repeatedly, so you can't easily determine how
long it spends in each node"
Sorry, looks like my prior statement (#2) was misleading. Actually, I don't
care so much about timing each condition of a rule but are looking to get
the total time it takes to evaluate all conditions of a rule. I am not
looking to break out statistics for each conditional statement of the rule.
Would this be easier/possible to do or does it face the same issues?
Thanks so much...was good talking to you guys at Rules Fest.
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Performance-monitorin...
Sent from the Drools - User mailing list archive at Nabble.com.
12:52 a.m.
On 07/11/2010 21:59, H.C. wrote:
Mark,
Thanks for your reply. I figured 3 and 4 were currently possible but 1 and 2
puzzled me.
"it interates from root to tip repeatedly, so you can't easily determine how
long it spends in each node"
Sorry, looks like my prior statement (#2) was misleading. Actually, I don't
care so much about timing each condition of a rule but are looking to get
the total time it takes to evaluate all conditions of a rule.
This isn't
possible as conditions are shared amoung rules and a rule's
conditions are not tested all at the time time, but incrementally as
knowledge is added that matches that rule. The best you can get is the
time it takes to evaluate all conditions for all rules by testing the
time it takes for an insert to do.
I am not
looking to break out statistics for each conditional statement of the rule.
Would this be easier/possible to do or does it face the same issues?
same issue.
Thanks so much...was good talking to you guys at Rules Fest.
11:36 a.m.
Thanks Mark, will try that.
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Performance-monitorin...
Sent from the Drools - User mailing list archive at Nabble.com.
5132
days inactive
5135
days old
4 comments
2 participants
participants (2)
-
H.C. -
Mark Proctor