Path: utzoo!mnetor!uunet!husc6!bbn!rochester!udel!burdvax!sdcrdcf!trwrb!aero!venera.isi.edu!raveling
From: raveling@vaxa.isi.edu (Paul Raveling)
Newsgroups: comp.software-eng
Subject: Re: A Cynic's Guide, part 1
Message-ID: <5071@venera.isi.edu>
Date: 17 Mar 88 17:28:29 GMT
References: <2541@Shasta.STANFORD.EDU> <5313@utah-cs.UUCP> <5019@venera.isi.edu> <5346@utah-cs.UUCP>
Sender: news@venera.isi.edu
Reply-To: raveling@vaxa.isi.edu.UUCP (Paul Raveling)
Distribution: na
Organization: USC-Information Sciences Institute
Lines: 50

In article <5346@utah-cs.UUCP> shebs%defun.utah.edu.UUCP@utah-cs.UUCP (Stanley T. Shebs) writes:
>
>This is the first I've heard of Unix being accused of slowing workstations
>to sub-PC levels!  Certainly doesn't correspond to my experience, where I've
>given up on PCs completely because they're just too excruciatingly slow to use,
>compared to the HP 350 on my desk...

	You're right that the 350 is good, fast machine.  That's
	what I'm using now, with 24 MB of RAM and almost 400 MB of disk.

	However, before that I used a 320.  The fellow who complained
	about his speed when compared to his PC's speed is also
	using a 320.  Both of us own AT clones, and we can both attest
	to the obvious speed difference.

	The context switch timing (Unix slower than at least one
	other system by an order of magnitude) is from benchmarks
	between EPOS and Unix V6 on the same PDP-11/45.  Some
	limited benchmarks on 350's, 320's,  and VAXes using
	HP-UX and BSD 4.2 suggested this ratio would hold true
	with the newer Unix systems.  I'd love to get a charter
	to write enough of an EPOS-like kernel to prove this on
	our workstations.

	Another benchmark of note was a FutureNet DASH-2/DASH-3 port.
	DASH-1 through DASH-4 were schematic designers carefully
	written in assembly language for PC's, including XT's.
	Before the original Bobcats (using 68010's) hit the streets
	a group ported DASH-2 to one, using C and Starbase to drive
	the graphics.  The Bobcat hardware was about 3 times faster
	than a PC/XT by most benchmarks.  The time required to
	display a particular drawing was 5 times longer on the Bobcat
	than on the XT, suggesting the software as a whole was
	15 times slower.

	Another case in point is our Geographic Display Agent on
	the HP workstations.  Its performance improved by an order
	of magnitude when I moved its X interface out of a separate
	graphics-subsystem process and into the GDA process.

	Finally, there are old benchmarks on VAX-11/780's comparing
	BSD 4.1 to VMS Version 2.  I don't recall the exact numbers
	now, but I believe they showed a context switch time of
	about 175 microseconds on VMS and about 1 millisecond on BSD.



---------------------
Paul Raveling
Raveling@vaxa.isi.edu