Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!yale!quasi-eli!cs.yale.edu!ciancarini-paolo
From: ciancarini-paolo@cs.yale.edu (paolo ciancarini)
Newsgroups: comp.lang.prolog
Subject: Re: Demons in Prolog
Summary: We had some experiences in Shared Prolog
Message-ID: <27563@cs.yale.edu>
Date: 3 Dec 90 21:04:13 GMT
References: <4716@mahendo.Jpl.Nasa.Gov>
Sender: news@cs.yale.edu
Organization: Yale University Computer Science Dept., New Haven, CT 06520-2158
Lines: 56
Nntp-Posting-Host: poe.systemsy.cs.yale.edu
Originator: cianca@poe.CS.Yale.Edu

In article <4716@mahendo.Jpl.Nasa.Gov> alan@nereid.jpl.nasa.gov (Alan Quan) writes:
>
>
>I need to find out if it's possible to implement demons in Prolog in an
>efficient manner.  These can be simple triggers which respond to
>pre-specified situations in the input data and then initiate a standard
>backward-chained Prolog inference (I'm also interested in efficient
>Prolog forward chaining in general, but this is not essential).
>

In the current implementation
of Shared Prolog, a distributed logic language defined and implemented 
by myself and other people at University of Pisa, Italy,
we use both backward and forward chaining as a technique to implement
parallel logic agents and triggers.
An SP program defines a set of logic agents that cooperate across a blackboard,
i.e. a dynamic knowledge base that is actually a Tuple Space a la Linda. 
Communication operations are in fact in(Atom) and out(Atom), that have a semantic
similar to assert and retract (I am afraid that this rule out real-time efficiency,
but actually our performance experiences are quite encouraging; and for the moment
we are not using Linda for the communication kernel).

Agents are triggered by patterns that have the following structure:

Read In | Goal Out

where Read and In are logic goals that define an activation guard (a trigger) that
when satisfied with respect the current contents 
of the blackboard starts a local Prolog goal
(Shared Prolog is an extension of Prolog); when this terminates
the Out is written back to the blackboard.
So each agent uses locally backward chaining, while
the blackboard as a whole can be seen as a working memory
of a logic bases production system (that uses forward chaining).
We could also define global distributed backtracking in a particular
application, but of course it is very computationally expensive,
so this is not an embedded linguistic mechanism (as opposed to
DeltaProlog, that I believe in its last version included
distributed backtracking as a linguistic feature).

We implemented Shared Prolog on a network of non homogeneous workstations
(Vaxes and Suns).

Shared Prolog is a project in progress. It is used to design a complex
programming environment.

Some formal details are contained in a paper forthcoming on 
ACM TOPLAS (January 91).
Some implementation details are contained
in a paper published in Proc. ACM Symp on Principles and Practice of Parallel Programming,
Seattle 1990, pp.40-49.

Paolo Ciancarini
Visiting Scientist
Dept. of Computer Science
Yale University


Brought to you by Super Global Mega Corp .com