Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/12/84; site desint.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!gamma!epsilon!zeta!sabre!petrus!bellcore!decvax!ittatc!dcdwest!sdcsvax!sdcrdcf!trwrb!desint!geoff
From: geoff@desint.UUCP (Geoff Kuenning)
Newsgroups: net.unix,net.unix-wizards
Subject: Re: Unix on top of/in parallel with other operating systems
Message-ID: <170@desint.UUCP>
Date: Wed, 5-Mar-86 17:24:47 EST
Article-I.D.: desint.170
Posted: Wed Mar  5 17:24:47 1986
Date-Received: Sat, 8-Mar-86 02:34:39 EST
References: <172@cybavax.UUCP> <193@dg_rtp.UUCP> <3310@sun.uucp>
Reply-To: geoff@desint.UUCP (Geoff Kuenning)
Distribution: net
Organization: SAH Consulting, Manhattan Beach, CA
Lines: 21
Xref: watmath net.unix:7315 net.unix-wizards:17097

In article <3310@sun.uucp> guy@sun.uucp (Guy Harris) writes:

> Huh?  For CPU-bound jobs, the only difference should be in OS overhead.  If
> you have only one process running, UNIX definitely doesn't impose much
> overhead and I'd be very surprised if AOS/*VS imposed much either.  There
> may be more overhead in context-switching and the like.  There may be more
> overhead in doing the I/O to read or write the data in under UNIX, I dunno.

An OS can slow a program by paging poorly (thrashing), and by swapping
poorly (which is different from paging poorly).  Also, most real large
computations are run under multitasking operating systems (you can't afford
to run dedicated), so the quality of the CPU scheduler, in terms of absolute
overhead, CPU thrashing characteristics, and throughput, matters a bunch.
UNIX, of course, falls down on almost every one of these.  The exceptions are
CPU thrashing (a one-second timeslice *can't* thrash a CPU bigger than
a 4004) and absolute overhead (at least in the more recent USG and BSD
systems where they got rid of all those linear searches).
-- 

	Geoff Kuenning
	{hplabs,ihnp4}!trwrb!desint!geoff