Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!wuarchive!gem.mps.ohio-state.edu!tut.cis.ohio-state.edu!bloom-beacon!eru!luth!sunic!mcsun!inria!irisa!priol
From: priol@irisa.irisa.fr (Thierry Priol,TB131,Equipe Hypercubes,9936200-547,)
Newsgroups: comp.arch
Subject: Re: parallel systems
Message-ID: <1646@irisa.irisa.fr>
Date: 25 Oct 89 08:12:34 GMT
References: <1989Oct24.210602.2616@ux1.cso.uiuc.edu>
Sender: news@irisa.irisa.fr
Lines: 23

A software user point of view:

Sometimes, it is interesting to simulate a shared memory on a
distributed memory parallel computers. We have found that a
ray-tracing algorithm is more efficient on a DMPC when it
is implementing with a shared memory programming model. It's 
due by the fact that it is difficult to partition the database
a priori (for example with a geometric partition). A software
shared memory service makes a dynamic data distribution (when
caches are using) during the computation. A paper which describes
our experimentation will be available soon. May be other algorithms
are in the same case ?

I hope that such a software shared memory service would be available
on DMPC in future operating system.

Thierry PRIOL

PS: I apologize for my poor english.

---------------------------------------------------------------------
Thierry PRIOL                                Phone:  99 36 20 00
IRISA / INRIA U.R. Rennes                    Fax:    99 38 38 32