Path: utzoo!attcan!uunet!lll-winken!lll-tis!ames!xanth!kent
From: kent@xanth.cs.odu.edu (Kent Paul Dolan)
Newsgroups: comp.sys.amiga
Subject: Re: Amiga UNIX
Message-ID: <5495@xanth.cs.odu.edu>
Date: 8 Jun 88 22:30:47 GMT
References: <211@laic.UUCP> <3663@cbmvax.UUCP> <1872@sugar.UUCP> <134@ssdis.UUCP> <2078@sugar.UUCP> <217@toylnd.UUCP>
Reply-To: kent@xanth.UUCP (Kent Paul Dolan)
Organization: Old Dominion University, Norfolk Va.
Lines: 24


While the whole discussion is sort of vaporous anyway, I hear folks on opposite
sides of the VM question talking past each other, so this is just a note to
inject some commonality into the discussion.

On a big, multiuser machine, Virtual Memory is a big win, because the
swapping time is used by other processes needing time, and the lossage
is limited to process switch times, page selection overhead, and
perhaps some DMA cycle stealing as well.

On a single user machine, all the swap time, if only a single task is
executing, is direct lossage, plus all the overhead losses.  Because
of the slow hard disk transfer rates, there can be a significant
slowdown perceptible to the user from choosing VM over added memory.
Now choosing an efficient size for a working set of pages, and
assuring by the choice of page sizes that the page fault rate is very
low are made even more important than to the designer of the big
multiuser machine.

The other answer, of course, is a low priority ray tracer running on
some hidden screen, so that even though the job in front of us runs
considerably more slowly, we can smirk that we are putting all those
excess cycles to use, unlike _some_ people we know.

Kent, the man from xanth.