[rules-users] drools performance with <

Fri Jan 22 09:51:22 EST 2010

   Tom,

   The way to improve performance is to break the combinatorial problem by
using one or both of the following:

* Write patterns against different classes: instead of writing the rule as
A() and A(), use different classes to represent each component and write A()
and B().

* Write alpha constraints (constraints against constant values) besides the
beta constraints (join). Example:

$a1: A( type == "source" )
$a2: A( type == "destination", someAttr > $a1.someAttr )

    Both of the above tricks help to break the worst case scenario ( O(n^p),
where n is the number of facts and p is the number of patterns).

    Edson

2010/1/21 Tom Carchrae <tom at carchrae.net>

> Please don't confuse my message with bashing Drools for not being as
> fast as Java.  While I am not very familiar with Rete, I'm aware of
> what goes on behind the scenes; I've been programming declarative
> optimization systems for many years.  My query relates to how to
> perform such a comparison with reasonable efficiency in Drools.
>
> Tom
>
>
> On Thu, Jan 21, 2010 at 7:16 PM, Greg Barton <greg_barton at yahoo.com>
> wrote:
> > pit->out
> >
> > That falls into a rare class: two letter typos that pass a spell checker.
> :)
> >
> > --- On Thu, 1/21/10, Greg Barton <greg_barton at yahoo.com> wrote:
> >
> >> From: Greg Barton <greg_barton at yahoo.com>
> >> Subject: Re: [rules-users] drools performance with <
> >> To: "Rules Users List" <rules-users at lists.jboss.org>
> >> Date: Thursday, January 21, 2010, 9:04 PM
> >> This is not at all surprising.
> >> To set up and run the rules there's quite a bit of code
> >> running behind the scenes besides just the comparison test.
> >> (For instance, there's object creation just to track the
> >> fact that two objects matched the rule, and in this case
> >> those objects are created for every possible combination of
> >> TestObject instances before the first rule ever fires, then
> >> added to lists, and so on...  So in the case of 1000
> >> TestObject instances, there's 1,000,000 object creations and
> >> list additions just for that.)
> >>
> >> So, as has been said many times on this list, and probably
> >> in a few blogs, for any given algorithm a hard coded java
> >> version will invariably be faster than rules.  What
> >> rules gives you is a way to express solutions to complex
> >> problems in a flexible and comprehensible way.  It
> >> shines when the problems are so complex that coding them
> >> procedurally is pit of the question.
> >>
> >> The same relationship is there between C and java.
> >> Java gives you managed, portable code, but for that you give
> >> up the raw speed of C.
> >>
> >> --- On Thu, 1/21/10, Tom Carchrae <carchrae at gmail.com>
> >> wrote:
> >>
> >> > From: Tom Carchrae <carchrae at gmail.com>
> >> > Subject: [rules-users] drools performance with <
> >> > To: rules-users at lists.jboss.org
> >> > Date: Thursday, January 21, 2010, 6:40 PM
> >> > Hello,
> >> >
> >> > I'm new to Drools and am trying it out.  I've found
> >> it
> >> > easy enough so
> >> > far, but quickly ran into a performance bottleneck.
> >> > (perhaps an all
> >> > too familiar story! :) ).  So, I made a simple
> >> example
> >> > to try and
> >> > track down the behavior.  Here it is.
> >> >
> >> > I wanted to create a rule that uses an <
> >> comparison
> >> > between pairs of
> >> > objects.  The following are the performance results:
> >> >
> >> > 10 objects.  Java comparison took 0 ms
> >> > 10 objects.  Rule firing took 2516 ms
> >> > 100 objects.  Java comparison took 0 ms
> >> > 100 objects.  Rule firing took 1984 ms
> >> > 1000 objects.  Java comparison took 47 ms
> >> > 1000 objects.  Rule firing took 132500 ms
> >> > 10000 objects.  Java comparison took 4641 ms
> >> > 10000 objects. Rule firing took .... (please wait...)
> >> >
> >> > Here is the rule
> >> >
> >> > rule "A < B"
> >> >     when
> >> >         a : TestObject(
> >> > )
> >> >         b : TestObject(
> >> > valueDouble < a.valueDouble )
> >> >     then
> >> >
> >> > b.setValueint(b.getValueint()+1);
> >> > end
> >> >
> >> > Now, I do understand that this (worst-case) is a
> >> comparison
> >> > of N x N
> >> > objects... so, 1000 = 1000000 comparisons.  However,
> >> > ~100ms per
> >> > comparison seems relatively expensive for such a
> >> simple
> >> > check.  I am
> >> > using StatefulKnowledgeSession (I tried Stateless but
> >> this
> >> > gave only a
> >> > mild performance improvement - and I would also like
> >> to use
> >> > insert in
> >> > some of my rules).
> >> >
> >> > How would one speed up such a query?  I know part of
> >> > the problem is
> >> > the < comparison.  The results with == are much
> >> > better (presumably, it
> >> > can do lots of pruning of the space)
> >> >
> >> > Model creation 0
> >> > 10 objects.  Rule firing took 2829 ms
> >> > 100 objects.  Rule firing took 93 ms
> >> > 1000 objects.  Rule firing took 657 ms
> >> > 10000 objects.  Rule firing took 3375 ms
> >> > 100000 objects.  Rule firing took 33266 ms
> >> >
> >> > I have thought about clever tricks that I could do to
> >> try
> >> > and speed
> >> > things up, like sorting into groups, etc.  But it
> >> all
> >> > seems relatively
> >> > convoluted.  Am I correct in coming to the
> >> conclusion
> >> > that a <
> >> > comparison is not the kind of thing that you should do
> >> in
> >> > Drools?
> >> >
> >> > How would you go about this?  I would prefer to keep
> >> > all the logic in
> >> > the DRL files.  I'm guessing you do a Java based
> >> > comparison and use a
> >> > fire-once rule to invoke this code.
> >> >
> >> > Thanks in advance for any insight,
> >> >
> >> > Tom
> >> > _______________________________________________
> >> > rules-users mailing list
> >> > rules-users at lists.jboss.org
> >> > https://lists.jboss.org/mailman/listinfo/rules-users
> >> >
> >>
> >>
> >>
> >>
> >> _______________________________________________
> >> rules-users mailing list
> >> rules-users at lists.jboss.org
> >> https://lists.jboss.org/mailman/listinfo/rules-users
> >>
> >
> >
> >
> >
> > _______________________________________________
> > rules-users mailing list
> > rules-users at lists.jboss.org
> > https://lists.jboss.org/mailman/listinfo/rules-users
> >
>
> _______________________________________________
> rules-users mailing list
> rules-users at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-users
>

-- 
 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/20100122/505e3115/attachment.html 