I have nothing empirical. Peter Lin recently ran the tests and said we
should be potentially 10-20 times faster than Drools 2.x - we are also
faster than Clips. However Jess is an awesome engine, if you can handle
lisp, and is still the speed demon - it beats many other commecial
engines. But remember Jess is not open source nor is it free, it is free
to trial, it is still very expensive to use commercially - typcailly 50K
USD or more for a large company.
1. What are you trying to empirically show here? The number of rules
executed is always the same
2. No idea - measing rulebase size is not normally a concern of rule
engines - unless yo are working in the embedded space. Its the size of
the working memory that is more an issue - I haven't measured this.
3. Again this question doesn't really make sense to me. Basically stay
away from evals and keep to nice literal and variable constraints. Evals
should only be used when a predicate's value change over time and is
thus not indexeable.
The article by Andrea is silly, the only thing that guy does is
demonstrate he knows little about rule engines.
Charles Young's article is more interesting. Although it compares Drools
2.x, which has known performance limits. It doesn't compare 3.0, which
is vastly more powerful than 2.x
Mark
Mark
Ho, Alan wrote:
Hi Mark,
I'm giving a presentation to my company tommorow on drools, and one question
that will come up will be the performance of drools. I know that the
performance is dependent on the type of rules, but I would like to know the
following :
1. How many rules does the manners64 execute ?
2. How big was the fact base ?
3. What kind of rules allows drools to scale linearly between time and the
number of rules ?
I'm sorry that this is of such short notice.
Here is some stuff that I saw :
http://weblogs.java.net/blog/schaefa/archive/2005/10/drools_performa.html
http://geekswithblogs.net/cyoung/articles/54022.aspx
Regards,
Alan Ho
-----Original Message-----
From: Mark Proctor [mailto:[EMAIL PROTECTED]
Sent: Thursday, March 09, 2006 7:06 AM
To: dev@drools.codehaus.org; Drools User List
Subject: [drools-user] JESS6 vs Drools3 vs CLIPS 6.23
http://woolfel.blogspot.com/2006/03/jess6-vs-drools3-vs-clips-623.html
Since last Spring, Drools has been making steady progress on improving the core
RETE implementation. Tonight I ran manners64 in both Jess6 and Drools3.
Jess 6 - 2208 ms
Drools 3 - 9781 ms
CLIPS 6.23 - 67.219 seconds
To put things in perspective, JESS is one of the fastest RETE engines and many
commercial engines can't beat it. Take for example Blaze Advisor, which has
been around for over a decade. The old blaze engine was slow. It's well known
within the rule industry. Last year FairIssacs bought RulesPower, to get a
faster engine.
In the last 8 months, Mark has rewritten Drools core and managed to make
significant performance and scalability improvements. Using the old
Drools2 core, 16 guests took several minute. Now 64 guests finishes within 10
seconds. A modest guess is that drools3 is atleast 10-20x faster than drools2.
Hopefully the users will appreciate all the hard work the drools team put into
drools3. It's nice to know the code I contributed last year has helped Drools 3
match CLIPS 6.23 performance.