Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wuarchive!uunet!stanford.edu!neon.Stanford.EDU!news
From: philip@pescadero.stanford.edu (Philip Machanick)
Newsgroups: comp.sys.mac.system
Subject: Re: Virtual Memory and Sys 7
Message-ID: <1991May9.172816.5477@neon.Stanford.EDU>
Date: 9 May 91 17:28:16 GMT
References: <1991May8.143042.20137@bigsur.uucp> <1991May9.042425.598@gla-aux.uucp> <1991May9.162601.19210@mintaka.lcs.mit.edu>
Sender: news@neon.Stanford.EDU (USENET News System)
Organization: Computer Science Department, Stanford University
Lines: 34

In article <1991May9.162601.19210@mintaka.lcs.mit.edu>
dbert@mole.gnu.ai.mit.edu (Douglas Siebert) writes:
>Virtual memory provides a temporary solution, but the way things are going,
>I'll wager virtual memory will not be used ten years from now.  Because if
>memory usage continues to double, we'll be running around 16G (that's 16,384M
>for those of you not aware what a "G" is :) )  Of course we'll have run into
>the 680x0's addressable limit of 4G by then, but many of us will have migrated
>to 64-bit processors by then.
>
>Anyway, if we have even 2G of physical RAM by then (a conservative estimate
I'd
>bet) we'll need to allocate several GIGS of memory on our harddisks to act as
>virtual memory.  Now even if hard disk technology has progressed as well as it
>has in the past 10 years (from 10M winchesters to 1G Fujitsus) I'd hate to 
>think how *slow* things would be, if we try to use much of that virtual
memory.

You raise some interesting questions. I don't think they mean the end for VM,
but rather the need to reimplement it. For example, one idea being worked on by
VM researchers is the notion that it may be faster to recompute some data than
to page it in and out, so the paging system gives the program the option of
keeping the data or not when it wants to replace pages.

Also, there is serious work on getting higher speed out of disks. Disk arrays
give you much faster transfer by putting several disks in parallel.

Another point: more mulitasking may mask the cost of paging, by having more
alternative processes to schedule when a page fault occurs.

It is amazing that the Mac OS has lasted as well as it has (even if it's
starting to look like a case of the 700-year old axe which has had 3 new blades
and 2 new handles).

Philip Machanick