Path: utzoo!attcan!uunet!husc6!babbage!reiter
From: reiter@babbage.harvard.edu (Ehud Reiter)
Newsgroups: comp.arch
Subject: Killer Micros
Message-ID: <3257@husc6.harvard.edu>
Date: 25 Nov 89 17:23:10 GMT
Sender: news@husc6.harvard.edu
Reply-To: reiter@harvard.harvard.edu ()
Organization: Aiken Computation Lab Harvard, Cambridge, MA
Lines: 30

Two comments on Killer Micros vs Mainframes:

(1) For large systems, computer cost is often just a small part of total
cost.  Suppose, say, that corporate buyer X noticed that a $100K Killer Micro
had equivalent compute power to the $1M mainframe he was considering.
This sounds like a dramatic savings.  But, when the buyer adds in the $5M
he's going to spend on peripherals, software, building costs, etc, he's
looking at a system cost of $5.1M for the Killer Micros, vs $6M for the
mainframe.  It's a savings, but not nearly as dramatic a one - and considering
that a bad computer system could cost his company $10M+ in lost business,
one could forgive the buyer for being cautious.

(2) Mainframe operating systems (e.g. MVS) are much superior to Killer Micro
OS's (inevitably a UNIX variant) in terms of reliability, security, error
recovery and the like, and also make far better use of that $5M worth of
peripherals.  Sure, programming under MVS is a nightmare (I've done it),
and most programmers would probably prefer to write UNIX code.  But, from
an end-user's perspective, MVS has a lot of advantages, and the above-mentioned
corporate buyer is unlikely to move to a Killer Micro system until such a
system has a competitive OS.


Killer Micros will make headway, but it will be slow headway.  It's taken
DEC 20 years to go from supplying well-regarded small scientific
computers (e.g. PDP-8's) to *starting* to supply corporate data centers,
and I suspect Killer Micros face the same kind of timescale.

					Ehud Reiter
					reiter@harvard	(ARPA,BITNET,UUCP)
					reiter@harvard.harvard.EDU  (new ARPA)