Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!rutgers!cbmvax!snark!eric
From: eric@snark.UUCP (Eric S. Raymond)
Newsgroups: comp.sys.ibm.pc,comp.arch,comp.unix.wizards
Subject: super new UNIX machine -- but are the disks fast enough?
Message-ID: <125@snark.UUCP>
Date: Sat, 15-Aug-87 00:40:08 EDT
Article-I.D.: snark.125
Posted: Sat Aug 15 00:40:08 1987
Date-Received: Thu, 20-Aug-87 04:44:24 EDT
Organization: Thyrsus Enterprises, Malvern PA 19355
Lines: 54
Xref: mnetor comp.sys.ibm.pc:6857 comp.arch:1830 comp.unix.wizards:3759

Ny new personal micro is a Compaq-386 clone running at 16 Mhz. The memory
is 100-nsec static-column DRAMs, so it runs at something between 0 and 1
wait states. The box consequently goes like bat out of hell on those in-memory
straightaways, even compared to my 3B1 which (using a 10Mhz 68010) is no
slouch. A copy of Microport's 386 System V.3 with DOS bridge is on order for
it.

Now for the bad news. My mass storage devices are a pair of Seagate 251s
hanging off a Western Digital WA-2 ST506 controller, rated at (I think) 80
milliseconds average access time (someone please correct me if that's wrong).
My questions:

	1) How can I tell if I've fallen into the minicomputer trap? -- that
	   is, how can I tell if my I/O bandwidth is low enough that I'll
	   mostly lose my processor speed advantage because I'm waiting on
	   disk transfers?

	2) If (1) is a problem, is there some method that can give me a rough
	   idea of what kind of drive performance I should pay for to achieve
	   a given fraction of maximum theoretical performance for the
	   processor-memory combination?

	3) I've read man pages for system activity profilers and I know how
	   to get down and dirty with the System V tunable parameter tables.
	   I don't yet have a good feel for how to hunt systematically for
	   performance/economy maxima in that parameter space. Comments and
	   war stories from unix.wizards with performance tuning savvy are
	   solicited.

I've cross-posted this to comp.arch because the situation has aroused my
curiosity about general methods for matching I/O subsystem and processor
performance -- is this still a black art?


For those of you looking for a good cheap expandable UNIX box:

The system cost me $2495; you can get one like it for no more than $2695.
That includes 1 meg of RAM, the Winchesters and an AT-format 1.2 meg, plus a
MCA/HGA CRT controller (with printer port), and choice of Phoenix or Award
AT-compatible BIOS, all sitting in an AT-sized case.

The box is manufactured by an outfit called Micronics, but they don't sell
Q1; I got mine from an outfit called Vesta Computers (phone: 1-(800)-843-5278).
The Vesta people and the Micronics tech I talked with were open, helpful and
a pleasure to deal with, and Vesta's prices are very aggressive and dropping
weekly. Tell them Eric sent you.


Now if only the processor weren't an Intel chip... ;-( ;-)

-- 
      Eric S. Raymond
      UUCP:  {{seismo,ihnp4,rutgers}!cbmvax,sdcrdcf!burdvax,vu-vlsi}!snark!eric
      Post:  22 South Warren Avenue, Malvern, PA 19355    Phone: (215)-296-5718