On Oct 26, 2007, at 3:03 PM, Jinghai Rao wrote:
Hi Ernest,
Thank you so much for your rapid reply. I have tried what you
proposed and it really significantly reduces the memory usage.
A problem is that we can't use suggestion 3 and 4 in our system
because such error:
[java] Message: Can't use funcalls in backchained patterns
MAIN::dynamic_triple.
Our system has to use backward-chained template here. Do you have
any suggestion on possible alternatives?
No, not really. You could create the list and store it in a
defglobal, I suppose.
By the way, do you have any comprehensive document about the
performance issue? I have read the manual and Jess in action book,
but both are talking of this very briefly.
The FAQ describes a specific rule that creates too many partial
matches, and how to fix it; that's the most important thing to
understand.
I would also like to read your article about the OAV triples.
No article, just an occasional brief statement. If you read the other
material, this really should just follow.
Thank you so much.
Jinghai
On 10/26/07, Ernest Friedman-Hill <[EMAIL PROTECTED]> wrote: Hi,
There are some very serious issues here that will effect both
performance and memory usage. Fixing them should substantially reduce
your memory footprint.
First, I've have gone on record many times explaining why using
generic OAV triples like this is a bad architectural choice. You're
basically denying Jess some of its ability to downselect collections
of working memory elements by hiding useful information. This forces
larger indices to be built than would otherwise be needed, and that's
one irreducible source of the large memory consumption here. I'll
ignore this though, as changing it would presumably require you to
rethink your whole project.
Second, note that the "or" conditional element should only be use to
express alternatives that actually differ in some way -- otherwise
you're just building a more complex Rete network for no reason, which
will, again, use more memory and run more slowly. The first "or"
below could simply be written
(need-authorized_triple
(predicate " http://mycampus.cs.cmu.edu/ontology#has_location";)
(subject ?owner)
(object ?location)
because ?owner and ?location will be bound to nil if the respective
slots contain nil; the explicit nil matches accomplish nothing that
the first pattern doesn't accomplish by itself, except expand the
size of the Rete network and its memory consumption for this rule by
a factor of four.
Third, the "or" conditional element should not be used to express
alternatives that differ only in the value of one slot in one pattern
-- that's what the "or" connective is for. So, for example, the
second "or" could be written more succinctly and more efficiently as
(and
(dynamic_triple
(predicate "http://www.w3.org/1999/02/22-rdf-syntax-ns#type ")
(subject " http://www-2.cs.cmu.edu/Group/rule1192814601";)
(object " http://mycampus.cs.cmu.edu/ontology#Group ")
)
(dynamic_triple
(predicate "http://mycampus.cs.cmu.edu/ontology#include";)
(subject " http://www-2.cs.cmu.edu/Group/rule1192814601 ")
(object ?sender | "http://www-2.cs.cmu.edu/People/Anybody";)
)
)
Fourth, you unify a bunch of facts on the "?day" variable, and then
later downselect by restricting ?day to one of four values. This
means you'll build a potentially huge set of partial matches, only to
throw some large number of them away. The test to restrict the value
of ?day should go at ?day's first use.
Finally, regarding how that ?day test is done: doing something like
this:
(and
(dynamic_triple
(predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type";)
(subject ?day)
(object " http://mycampus.cs.cmu.edu/QOWL#Day";)
)
(test (member$ ?day (create$ " http://mycampus.cs.cmu.edu/
day#Tuesday " " http://mycampus.cs.cmu.edu/day#Wednesday"; "http://
mycampus.cs.cmu.edu/day#Thursday" " http://mycampus.cs.cmu.edu/
day#Saturday" )))
)
is going to create a new list of values to test each single fact
that's considered here -- a waste of resources. Further, in Jess 7.0
and earlier, it's going to waste memory building up unneeded partial
matches. You could just write this:
(and
(dynamic_triple
(predicate "http://www.w3.org/1999/02/22-rdf-syntax-ns#type";)
(subject ?day & :(or (eq ?day " http://mycampus.cs.cmu.edu/
day#Tuesday ")
(eq ?day "http://mycampus.cs.cmu.edu/
day#Wednesday")
(eq ?day " http://mycampus.cs.cmu.edu/
day#Thursday")
(eq ?day " http://mycampus.cs.cmu.edu/
day#Saturday")))
(object "http://mycampus.cs.cmu.edu/QOWL#Day";)
)
On Oct 25, 2007, at 9:21 PM, Jinghai Rao wrote:
> Hello everybody,
>
> We are using jess to develop an access control system. The access
> control rule has been translated to a jess rule as shown below. We
> notice that adding one of such rule consumes 500K - 1MB more
> memory. I wonder if it is normal. Do you have any suggestion to
> reduce the memory usage by either changing the rule format or
> changing jess/java VM configuration?
>
> Here is an example of such rule. dynamic_triple and
> authorized_triple use backward chaining.
>
> (defrule rule1192814601 (declare (salience 50))
>
> (or
>
> (need-authorized_triple
> (predicate " http://mycampus.cs.cmu.edu/ontology#has_location";)
> (subject ?owner)
> (object ?location)
> )
> (need-authorized_triple
> (predicate " http://mycampus.cs.cmu.edu/ontology#has_location";)
> (subject nil)
> (object ?location)
> )
> (need-authorized_triple
> (predicate " http://mycampus.cs.cmu.edu/ontology#has_location";)
> (subject ?owner)
> (object nil)
> )
> (need-authorized_triple
> (predicate " http://mycampus.cs.cmu.edu/ontology#has_location";)
> (subject nil)
> (object nil)
> )
>
> )
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type";)
> (subject ?owner)
> (object " http://mycampus.cs.cmu.edu/ontology#Person ")
> )
>
> (dynamic_triple
> (predicate "http://mycampus.cs.cmu.edu/ontology#has_location ")
> (subject ?owner)
> (object ?location)
>
> )
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type ")
> (subject ?location)
> (object "http://mycampus.cs.cmu.edu/ontology#Location ")
> )
>
> (and
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type ")
> (subject ?query)
> (object " http://mycampus.cs.cmu.edu/QOWL#Query";)
> )
>
> (dynamic_triple
> (predicate " http://mycampus.cs.cmu.edu/QOWL#querytime";)
> (subject ?query)
> (object ?querytime)
>
> )
>
> (dynamic_triple
> (predicate " http://mycampus.cs.cmu.edu/QOWL#gpstime";)
> (subject ?query)
> (object ?gpstime)
>
> )
>
> (dynamic_triple
> (predicate " http://mycampus.cs.cmu.edu/QOWL#sender";)
> (subject ?query)
> (object ?sender)
>
> )
>
> )
>
> (and
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type";)
> (subject ?querytime)
> (object " http://mycampus.cs.cmu.edu/QOWL#Datetime";)
> )
>
> (dynamic_triple
> (predicate " http://mycampus.cs.cmu.edu/QOWL#hasDay ")
> (subject ?querytime)
> (object ?day)
>
> )
>
> (dynamic_triple
> (predicate " http://mycampus.cs.cmu.edu/QOWL#hasTime ")
> (subject ?querytime)
> (object ?time)
>
> )
>
> )
>
> (and
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type";)
> (subject ?gpstime)
> (object " http://mycampus.cs.cmu.edu/QOWL#Datetime";)
> )
>
> (dynamic_triple
> (predicate "http://mycampus.cs.cmu.edu/QOWL#hasDay ")
> (subject ?gpstime)
> (object ?gpsday)
>
> )
>
> (dynamic_triple
> (predicate "http://mycampus.cs.cmu.edu/QOWL#hasTime ")
> (subject ?gpstime)
> (object ?gpshourmin)
>
> )
>
> )
>
> (or
>
> (and
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type ")
> (subject "http://www-2.cs.cmu.edu/Group/rule1192814601 ")
> (object " http://mycampus.cs.cmu.edu/ontology#Group";)
> )
>
> (dynamic_triple
> (predicate " http://mycampus.cs.cmu.edu/ontology#include";)
> (subject "http://www-2.cs.cmu.edu/Group/rule1192814601 ")
> (object ?sender)
>
> )
>
> )
>
> (and
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type ")
> (subject " http://www-2.cs.cmu.edu/Group/rule1192814601";)
> (object " http://mycampus.cs.cmu.edu/ontology#Group";)
> )
>
> (dynamic_triple
> (predicate " http://mycampus.cs.cmu.edu/ontology#include ")
> (subject "http://www-2.cs.cmu.edu/Group/rule1192814601 ")
> (object " http://www-2.cs.cmu.edu/People/Anybody";)
>
> )
>
> )
>
> )
>
> (or
>
> (and
>
> (and
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type";)
> (subject ?day)
> (object " http://mycampus.cs.cmu.edu/QOWL#Day ")
> )
>
> (test (member$ ?day (create$ "http://mycampus.cs.cmu.edu/
> day#Tuesday " " http://mycampus.cs.cmu.edu/day#Wednesday"; "http://
> mycampus.cs.cmu.edu/day#Thursday " " http://mycampus.cs.cmu.edu/
> day#Saturday" )))
>
> )
>
> (and
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type ")
> (subject ?time)
> (object "http://mycampus.cs.cmu.edu/QOWL#Time ")
> )
>
> (dynamic_triple
> (predicate " http://mycampus.cs.cmu.edu/ROWL#any";)
> (subject ?time)
> (object "false")
>
> )
>
> (test (> ?time "0400"))
>
> (test (< ?time "0700"))
>
> )
>
> )
>
> )
>
> (or
>
> (and
>
> (and
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type";)
> (subject ?gpsday)
> (object " http://mycampus.cs.cmu.edu/QOWL#Day ")
> )
>
> (test (member$ ?gpsday (create$ " http://mycampus.cs.cmu.edu/
> day#Tuesday" " http://mycampus.cs.cmu.edu/day#Wednesday"; "http://
> mycampus.cs.cmu.edu/day#Thursday " "http://mycampus.cs.cmu.edu/
> day#Saturday " )))
>
> )
>
> (and
>
> (dynamic_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type ")
> (subject ?gpshourmin)
> (object "http://mycampus.cs.cmu.edu/QOWL#Time ")
> )
>
> (dynamic_triple
> (predicate " http://mycampus.cs.cmu.edu/ROWL#any";)
> (subject ?gpshourmin)
> (object "false")
>
> )
>
> (test (> ?gpshourmin "0400"))
>
> (test (< ?gpshourmin "0700"))
>
> )
>
> )
>
> )
>
>
> =>
>
> (assert
> (authorized_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type ")
> (subject ?owner)
> (object " http://mycampus.cs.cmu.edu/ontology#Person";)
> )
> )
> (assert
> (authorized_triple
> (predicate " http://mycampus.cs.cmu.edu/ontology#has_location";)
> (subject ?owner)
> (object ?location)
>
> )
> )
> (assert
> (authorized_triple
> (predicate " http://www.w3.org/1999/02/22-rdf-syntax-ns#type";)
> (subject ?location)
> (object " http://mycampus.cs.cmu.edu/ontology#Location";)
> )
> )
> (add-to-log ruleID rule1192814601 conversationID ?query
> location ?location)
> )
>
>
> Thank you very much.
>
> Jinghai
>
>
>
>
>
> --
> ------------------------------------
> Jinghai Rao
> Project Scientist
>
> Institute for Software Research International
> School of Computer Science
> Carnegie Mellon University
> 5000 Forbes Avenue, 234 Smith Hall,
> Pittsburgh, PA 15213
>
> Telephone: +1 412 268 1357
> Fax: +1 412 268 7287
> Mobile: +1 412 580 9756
> http://www.cs.cmu.edu/~jinghai
---------------------------------------------------------
Ernest Friedman-Hill
Informatics & Decision Sciences Phone: (925) 294-2154
Sandia National Labs FAX: (925) 294-2234
PO Box 969, MS 9012 [EMAIL PROTECTED]
Livermore, CA 94550 http://www.jessrules.com
--------------------------------------------------------------------
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 owner-jess-
[EMAIL PROTECTED] .
--------------------------------------------------------------------
--
------------------------------------
Jinghai Rao
Project Scientist
Institute for Software Research International
School of Computer Science
Carnegie Mellon University
5000 Forbes Avenue, 234 Smith Hall,
Pittsburgh, PA 15213
Telephone: +1 412 268 1357
Fax: +1 412 268 7287
Mobile: +1 412 580 9756
http://www.cs.cmu.edu/~jinghai
---------------------------------------------------------
Ernest Friedman-Hill
Informatics & Decision Sciences Phone: (925) 294-2154
Sandia National Labs FAX: (925) 294-2234
PO Box 969, MS 9012 [EMAIL PROTECTED]
Livermore, CA 94550 http://www.jessrules.com
--------------------------------------------------------------------
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]
--------------------------------------------------------------------
