Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!zaphod.mps.ohio-state.edu!samsung!viewlog.viewlogic.com!peter
From: peter@suntan.viewlogic.com (Peter Colby)
Newsgroups: comp.sys.mac.system
Subject: Re: VM rule of thumb (sic)
Message-ID: <1991May31.135034.16813@viewlogic.com>
Date: 31 May 91 13:50:34 GMT
References: <1991May30.132055.23209@xn.ll.mit.edu> <14845@ector.cs.purdue.edu> <1991May30.193537.21239@m.cs.uiuc.edu>
Sender: news@viewlogic.com
Organization: Viewlogic Systems, Inc., Marlboro, MA
Lines: 47
Originator: peter@hari
Nntp-Posting-Host: hari


In article <1991May30.193537.21239@m.cs.uiuc.edu>, gillies@m.cs.uiuc.edu (Don Gillies) writes:
|> My own rule of thumb gleaned from Pilot (Xerox OS) is also 2:1.  Thus
|> this rule is not just a UNIX artifact.  I think that most reasonable
|> systems, when the software is packaged/tuned for VM, can present
|> double the amount of physical memory without thrashing.
|> 
|> 128Mb of virtual memory v.s. 8Mb of physical memory on a LISP machine
|> is because LISP is a joke.  Common Lisp has thousands of useless
|> functions that need to be available, and are eating that swap space.
|> You have other junk in your environment like a compiler or a debugger
|> that by all rights is wasting VM 100% of the time.  And the majority
|> of garbage collection algorithms are dogs if memory is more than 50%
|> occupied.
|> 
|> So while you *say* you have 128Mb of virtual memory, you are probably
|> *using* tons less memory, the rest being garbage of one sort or
|> another.  Too bad for you, great for the disk drive makers.

	I am the originator of this "VM Rule of thumb" thread. That being
stated, I have another comment regarding VM (or swap space).
	Yes, the UNIX (and apparently other systems) rule of thumb is 2:1
virtual to physical. However, I have worked on and seen a number of other
programs which NEED vast amounts of VM yet function adequately in fairly
small amounts of physical memory. Certain CAD and CAE programs are high
on the list. I would also imagine that complex statistical programs
(like Markov chain analysis) need lots of VM but work fine in small
amounts of physical.
	The issue here is locality of reference. If you have a program that
needs large arrays in memory then you need LOTS of swap; but if the program
only accesses small localized areas of those large data structures at any
particular time, then page faulting is minimal and a small physical memory
is adequate for the computations.
	Then again, these programs could do their OWN memory management and
work perfectly well without VM at all - as in fact happened before the days
of VM. I worked for a company that had a program that was written (and
overlayed for) a physical address space of 64K on a mini-computer. The
code space required for this program was about 8MB (yes, 2+ MILLION lines
of code). This massive program was ported to a VM workstation with a 2^32
address space and immediately required 50MB of swap. I believe the
original workstations we ported to could take somewhere between 8-16MB
max. There was a small loss in performance but not significant.
	P
-- 
      (O)(O)(O)(O)(O)(O)(O)(O)(O)     (O)(O)(O)(O)(O)(O)(O)(O)(O)
      (O) !the doctor is out! (O)     (0) peter@viewlogic.com (0)
      (O)(O)(O)(O)(O)(O)(O)(O)(O)     (O)(O)(O)(O)(O)(O)(O)(O)(O)