Path: utzoo!mnetor!uunet!husc6!mailrus!ames!pasteur!ucbvax!BU-CS.BU.EDU!bzs
From: bzs@BU-CS.BU.EDU (Barry Shein)
Newsgroups: comp.society.futures
Subject: What to do with all those MIPS
Message-ID: <8804191303.AA24772@bu-cs.bu.edu>
Date: 19 Apr 88 13:03:12 GMT
References: <3b8a861f.44e6@apollo.uucp>
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 62


Ah, my favorite subject...

I've been talking for a year or so with vendors like Encore who are
threatening (?) to deliver "minis" in the range of 1000MIPs or more in
the near future (12..24 months.) I suspect this might breathe new
interest into a currently rather boring mini market. One interesting
question also is what will some of the folks who build mainframes do
when all this happens (although they retain a big lead in disk i/o
performance some of that will surely falter also.) I mean, really,
deliver 10,000MIPS in the same time frame?

No, I don't think that will be the rational response to workstations
with 100MIPs (<$100K) and minis with 1000MIPs ($100K..$500K.) There
will be customers (JC Penneys) who just need the I/O channels of the
big mainframes and computes are secondary (an IBM3090/600 should
deliver around 180MIPs right now, that's not *that* shameful :-), but
I can't help but think there are a lot of customers out there who may
hesitate to blow $10M on a mainframe if they can do it on a $200K mini
(not to mention the power, real estate, operational philosophy [you
don't need 50 people to run a mini like you do a mainframe] etc.)
Seriously, how big can their jobs be? How much can they possibly have
grown since they managed on a 10MIPs mainframe 5 years ago?

Specifically, I have been interested in something I refer to as
"wasteful computing". We have reached a point where ther is probably
more "waste" from idle processors than over-utilization (or are about
to.) The software hasn't kept up (as was predicted by most everyone.)

Note that you should avoid the value-laden meaning of "waste" in
this context. There's nothing wrong with an idle CPU, it just is
interesting.

Modern workstations already exhibit the beginnings of wasteful
computing. If you showed those bitmap screens and the computations
involved in updating them to someone in the 60's they would have run
screaming out the door. Imagine paying for the cycles and
kilocore-ticks for dragging a window across the screen? Probably would
have cost you $15 or so in 1975 at government rates (they have std
rates is all I mean, form A21.)

Here's something specific that occurred to me the other day. Everyone
remember the "histogram" or "frequency" sort? That's a sort algorithm
that runs linear to N (N = # of elements.) It works like this:

	DECLARE BIGARRAY[ADDRESS_RANGE]

	while not EOF
	  READITEM(I)
	  BIGARRAY[I] = BIGARRAY[I] + 1
	end

That's it, when you hit EOF the data is sorted. It occurred to me that
you can now easily sort 24 bit data items on a modest workstation.
Most of this is due to larger memories and not MIPs but I think most
of you old timers would agree that you were taught this sort as
basically a useless algorithm due to its vast memory usage. Hmm,
seems useful again all of a sudden! What happened?!

Anyhow, more later...

	-Barry Shein, Boston University