Path: utzoo!utgpu!news-server.csri.toronto.edu!clyde.concordia.ca!uunet!cs.utexas.edu!asuvax!ncar!boulder!grunwald
From: grunwald@foobar.colorado.edu (Dirk Grunwald)
Newsgroups: comp.arch
Subject: Re: Shared Memory
Message-ID: <28071@boulder.Colorado.EDU>
Date: 13 Oct 90 07:00:12 GMT
References: <10678@pt.cs.cmu.edu> <1990Oct9.174307.27365@irisa.fr>
Sender: news@boulder.Colorado.EDU
Reply-To: grunwald@foobar.colorado.edu
Organization: University of Colorado at Boulder
Lines: 12
In-reply-to: priol@irisa.fr's message of 9 Oct 90 17:43:07 GMT


I'm not terribly suprised that a program with a largely read-only
database, such as a ray tracer, does better under SVM/NVM/DVM.

Consider such a program on a disk system w/o enough memory. You can
either check residence on each object access, or you can just have the
MMU handle it. The 95% case is that the object is local.

Of course, this is won't work so well if you naively transform a
program to a network memory machine; of course, porting a shared
memory problem to e.g, an XPC encore or the new BBN machines will be
hard if you don't think about it.