All:
In the past, I have worked with Advisor, JRules and OPSJ. I think if you will examine
the JRules "Optimized" version, you will find that the rules, priorities and data have
been arranged such that there is absolutely no time lost - basically it is a
procedural program that "expects" everything in the correct order. Also, JRULES-OPT
is not another rulebase; it is another ILR file in JRules. And it _should_ run
faster. It didn't have to "think" like the others did.
For example, in the file mannero.ilr, some of the rules use the following statement:
?C.assertPath ( ?path );
Because this is a Java routine, it does virtually nothing in working memory. However,
if this statement is modified so that it actually goes to working memory by restating
in something like:
modify ?C { assertPath ( ?path ) }
Then the "optimzed" version performs very similarly to the regular JRules version.
Another thing that I found interesting was that OPSJ ran even faster than the compiled
C/C++ rulebase engines, such as Haley and CLIPS and ILOG Rules, on the Manners 64 and
Manners 128. As I understand OPSJ, the reason for this is that OPSJ uses the Rete 2
algorithm which optimizes working memory efficiency as well as employing the Rete
network optimization.
BTW, which version of OPSJ were you using? Version 4 of OPSJ is supposed to be even
faster than Version 3.3, which is the last version that I tested last year some time.
One final point: I heard from Java One (Greg Barton) that JDK 1.3.1 is significantly
faster (about 200%) on Solaris than JDK 1.3.0. However, the improvement on NT was not
so drastic; only about 25% or so.
SDG
jco
-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]
Sent: Friday, June 29, 2001 1:09 PM
To: [EMAIL PROTECTED]
Cc: Al-Akhras, Khaled
Subject: RE: JESS: Manners test error
At 13:38 29/6/2001 -0400, Al-Akhras, Khaled wrote:
> How does it compare to the other implementations?
These are some of the data resulting from my tests:
Product Company 16 Guests 32 Guests 64 Guests 128 Guests
CLIPS NASA 0.4 1.5 52 2520
CLIPS\R2 PST 0.015 0.36 2.75 22.33
G2 Gensym 0.266 2.875 38.14 551.64
JESS 6.0 Sandia Labs 0.2 0.8 7.3 92
JRULES ILog 0.093 0.437 6.05 101.81
JRULES-OPT ILog 0.093 0.2 0.64 2.43
OPSJ PST 0.14 0.31 1.64 10.98
RETE++ Haley 0.03 0.34 3.46 63.75
RULES ILog 0.031 0.11 2.71 58.9
The time units are in seconds. Although not very relevant, the system used
was a Dell PowerEdge with Windows NT 4.0 Server, 256 MB RAM, and dual
750 MHz Pentium 3 processors. The Java based tests (Jess, JRules and
OPSJ) were also tried on a SunBlade 1000 with Solaris 8, 1GB RAM, dual
750 MHz UltraSPARC-III processors. It was slower in all the cases. In
both systems I used the same JDK 1.3.0.
Regards,
Juanjo
--
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+
| Juan Jos� Garc�a Adeva |
| Room 1G-628A mailto:[EMAIL PROTECTED] |
| Avaya Labs Research http://www.avayalabs.com|
| 101 Crawdords Corner Road Phone: +1-732-817-6228 |
| Holmdel, NJ 0733-3030 Fax: +1-732-817-4870 |
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+
---------------------------------------------------------------------
To unsubscribe, send the words 'unsubscribe jess-users [EMAIL PROTECTED]'
in the BODY of a message to [EMAIL PROTECTED], NOT to the
list (use your own address!) List problems? Notify [EMAIL PROTECTED]
---------------------------------------------------------------------
---------------------------------------------------------------------
To unsubscribe, send the words 'unsubscribe jess-users [EMAIL PROTECTED]'
in the BODY of a message to [EMAIL PROTECTED], NOT to the
list (use your own address!) List problems? Notify [EMAIL PROTECTED]
---------------------------------------------------------------------
