Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!spool2.mu.edu!sdd.hp.com!wuarchive!emory!hubcap!baude
From: baude@aar.alcatel-alsthom.fr (Francoise Baude)
Newsgroups: comp.parallel
Subject: Re: Is message passing more efficient than shared mem ?
Message-ID: <12721@hubcap.clemson.edu>
Date: 21 Jan 91 15:51:24 GMT
References: <12671@hubcap.clemson.edu>
Sender: fpst@hubcap.clemson.edu
Reply-To: baude@aar.alcatel-alsthom.fr (Francoise Baude)
Organization: Alcatel-Alsthom Recherche (Marcoussis)
Lines: 18
Approved: parallel@hubcap.clemson.edu

I have demonstrated that shared memory parallel programming model (represented by PRAMs)
is almost equivalent to the message passing parallel programming model (represented
by Actors) with the point of view of expressiveness and efficiency (product of 
parallel time and number of processors used).
'Almost' means that a O(log N) inneficiency factor, due to asynchronicity and 
distribution of message passing model, is unavoidable when you simulate a PRAM program 
by an Actor program. On the other side, the efficiency is constant when you simulate a 
PRAM program by an Actor program, because a O(log N) increase in parallel time is 
unavoidable but you can use less processors (in an order of magnitude of O(log N)).

The comparisons of the two kinds of models are made by assuming the SAME underlying
architecture (i.e. extendible, message passing, asynchronous, processors network) 
on which a unique universal 'Von Neuman' like execution model can be build for
any parallel programming model.

An article entitled 'Actors as a Parallel Programming Model' will appear in the 8th STACS proceedings, LNCS, in February 1990. If you are interested, I can send you a copy before the publication.

Francoise Baude 	email: baude@aar.alcatel-alsthom.fr