3:19 a.m.
Hi Drools team,
I wrote a Drools application for my company, which evaluates clinical claim
processing rules. That is, it checks current medical claims against the
patient's historical claims.
It appears that 95% of the run time is dedicated to inserting facts, which
is presumably when it propagates through the RETE graph, since firing the
rules takes no time.
But it seems like it's not fast enough. I made a rule template, which by
trial and error of moving the conditions around, is optimised. I'm
averaging 310ms, and need it to be under 170ms for it to be viable.
Is there any not-too-technical guide to optimisation somewhere? (alpha and
beta-nodes scare me a bit).
Regards,
Daniel
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Insert-Fact-Speed-bot...
Sent from the Drools - User mailing list archive at Nabble.com.
4 a.m.
@ djb ,
Can u tell me how did u inserted the facts urself ??
Is a just a simple assertion or any thing else ?
On 5/5/10, djb <dbrownell83(a)hotmail.com> wrote:
Hi Drools team,
I wrote a Drools application for my company, which evaluates clinical claim
processing rules. That is, it checks current medical claims against the
patient's historical claims.
It appears that 95% of the run time is dedicated to inserting facts, which
is presumably when it propagates through the RETE graph, since firing the
rules takes no time.
But it seems like it's not fast enough. I made a rule template, which by
trial and error of moving the conditions around, is optimised. I'm
averaging 310ms, and need it to be under 170ms for it to be viable.
Is there any not-too-technical guide to optimisation somewhere? (alpha and
beta-nodes scare me a bit).
Regards,
Daniel
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Insert-Fact-Speed-bot...
Sent from the Drools - User mailing list archive at Nabble.com.
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
5:07 a.m.
@puneet - I would suggest downloading the sample files, namely those that
come with the Drools plug-in for eclipse. I found the Conway Game of Life
example to be the most useful for learning past the basics. But your
question should be in another post.
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Insert-Fact-Speed-bot...
Sent from the Drools - User mailing list archive at Nabble.com.
4:12 a.m.
Not having any clue about the fact structure (field types) and rule patterns
makes it impossible to advise.
"Moving conditions around" is, of course, one way, but there might be
things you have overlooked; or essentially different techniques.
-W
On Wed, May 5, 2010 at 10:19 AM, djb <dbrownell83(a)hotmail.com> wrote:
Hi Drools team,
I wrote a Drools application for my company, which evaluates clinical claim
processing rules. That is, it checks current medical claims against the
patient's historical claims.
It appears that 95% of the run time is dedicated to inserting facts, which
is presumably when it propagates through the RETE graph, since firing the
rules takes no time.
But it seems like it's not fast enough. I made a rule template, which by
trial and error of moving the conditions around, is optimised. I'm
averaging 310ms, and need it to be under 170ms for it to be viable.
Is there any not-too-technical guide to optimisation somewhere? (alpha and
beta-nodes scare me a bit).
Regards,
Daniel
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Insert-Fact-Speed-bot...
Sent from the Drools - User mailing list archive at Nabble.com.
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
4:42 a.m.
Hi,
in your case is there no way to insert data at other moments ? do you need
all the data you insert ?
If you use statefull session you can insert data during time and not at
once.
Also, the more rules you have with complex when conditions => The rete is
big so inserting takes time.
On on of my project, using a rule flow makes it easier as you make decision
step by step. So for each step, the rete graph is smaller (it my view when
you divide in smaller steps) => Quicker when you insert.
Hope it helps
Nicolas Héron
2010/5/5 Wolfgang Laun <wolfgang.laun(a)gmail.com>
Not having any clue about the fact structure (field types) and rule
patterns
makes it impossible to advise.
"Moving conditions around" is, of course, one way, but there might be
things you have overlooked; or essentially different techniques.
-W
On Wed, May 5, 2010 at 10:19 AM, djb <dbrownell83(a)hotmail.com> wrote:
>
> Hi Drools team,
>
> I wrote a Drools application for my company, which evaluates clinical
claim
> processing rules. That is, it checks current medical claims against the
> patient's historical claims.
>
> It appears that 95% of the run time is dedicated to inserting facts,
which
> is presumably when it propagates through the RETE graph, since firing the
> rules takes no time.
>
> But it seems like it's not fast enough. I made a rule template, which by
> trial and error of moving the conditions around, is optimised. I'm
> averaging 310ms, and need it to be under 170ms for it to be viable.
>
> Is there any not-too-technical guide to optimisation somewhere? (alpha
and
> beta-nodes scare me a bit).
>
> Regards,
> Daniel
>
>
> --
> View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Insert-Fact-Speed-bot...
> Sent from the Drools - User mailing list archive at Nabble.com.
> _______________________________________________
> rules-users mailing list
> rules-users(a)lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-users
>
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
5 a.m.
Hi,
thanks for the replies. Essentially, its a rule template, with 4
conditions.
One checks for a current claimline,
One checks for a historic claimline,
One checks that the historic claimline is within a time period of the
current claimline,
One checks that there are X historic claimlines.
I'm just a bit hesitant to post my work here. But i did have a BIG
breakthrough.
What I was doing was Insert-Fire-Retract, once per claim. When I changed it
to Retract at the end, it jumped down to 50ms! (6X speed increase!)
But, unfortunately, i need to retract the data for each claim, because the
rules assume that everything in WM is relevant. (That is, it assumes all the
historic claims are for the current member)
Hmm. Any new ideas?
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Insert-Fact-Speed-bot...
Sent from the Drools - User mailing list archive at Nabble.com.
6:02 a.m.
Sorry, I re-read my post, and it might not make sense.
I changed from Insert-Fire-Retract in a loop (once per claim), to a loop of
Insert-Fire, then doing a bulk retract after a 100 or so claims. For some
reason, this decreased run time 600%
i think i'll modify the claims to have a context of who they belong to, and
then add an extra condition to link them. That is, if no one thinks perhaps
there is a way to keep the Insert-Fire-Retract loop and gain the same kind
of speed.
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Insert-Fact-Speed-bot...
Sent from the Drools - User mailing list archive at Nabble.com.
10:22 a.m.
On 05/05/2010 09:19, djb wrote:
Hi Drools team,
I wrote a Drools application for my company, which evaluates clinical claim
processing rules. That is, it checks current medical claims against the
patient's historical claims.
It appears that 95% of the run time is dedicated to inserting facts, which
is presumably when it propagates through the RETE graph, since firing the
rules takes no time.
But it seems like it's not fast enough. I made a rule template, which by
trial and error of moving the conditions around, is optimised. I'm
averaging 310ms, and need it to be under 170ms for it to be viable.
Is there any not-too-technical guide to optimisation somewhere? (alpha and
beta-nodes scare me a bit).
Other than writting the rules well - using == constraints, avoiding
cross products ete, there isn't much more you can do. If you have time
you can work with us on the next generation engine, which will emit
compiled java code, should be about 3 to 5 times faster. But the core
team aren't working on that quite yet, just ideas in our head.
Mark
Regards,
Daniel
3:40 a.m.
Thanks Mark,
I don't have much free time to contribute, unfortunately, (I've gotta look
busy at work...). sorry to be an abuser of open source :) But I'll
download the source, and occasionally try understand what's going on in the
dev forum.
Regards
Daniel
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Insert-Fact-Speed-bot...
Sent from the Drools - User mailing list archive at Nabble.com.
5314
days inactive
5319
days old
8 comments
5 participants
participants (5)
-
djb -
Mark Proctor -
Nicolas Héron -
Puneet duggal -
Wolfgang Laun