Path: utzoo!attcan!uunet!lll-winken!lll-tis!helios.ee.lbl.gov!pasteur!ucbvax!hplabs!hpda!hptsug2!taylor
From: eugene@eos.arc.nasa.gov (Eugene Miya)
Newsgroups: comp.society
Subject: Re: "Personal" Computers
Message-ID: <488@hptsug2.HP.COM>
Date: 31 Aug 88 17:22:11 GMT
Sender: taylor@hptsug2.HP.COM
Organization: NASA Ames Research Center, Calif.
Lines: 52
Approved: taylor@hplabs

While I like all the defense for personal computers, I personally find
the vision expressed in the three postings I have received as a little
disappointing.  Sorry, don't take it too personally.  First, off, it's
not clear to me that we will continue to have increasingly faster cycle
times.  100 Mz, 500 MHz, do you know what's going to bound this?  Can I
just add zeros (I doubt it)?

The next obvious thing is parallelism.  (great buzzword, used by
numerous people who know thing about it, before you use it you should
read The Mythical Man-Month and explain the concept of what a Mythical
MIPS might mean).  Don't get me wrong, I strongly endorse all the
research on what Gordon Bell calls Multis (Sequents (we have 3), Encores
(1), Flex (JSC and LaRC each have one), and so forths.  If you want an
impressive demo, I have one which forks 10 real processes working on
indexing my parallel processing bibliography (a real use).  The problem
is that there are still big hurdles.  If it were easy, it would have
been before.

Now, don't get me wrong.  During the history of personal workstations
conference, we received out first Cray-2 [serial 2003] ($17M, please,
this isn't meant to be a brag), and I needed some standalone time for
research, so I left the conference, got my hour, and returned.  It was
mind boggling to me.  Here I was, running the "future" Apple 700 (should
I use Roman numerals?)  Anyway, I think it's going to be EXTREMELY
important for users to get standalone time on fast machines as if they
were personal computers (this is the more of the same direction of
improvement, fine).  This will really help develop the software,
interaction is where LLNL and LANL got it right.  (sure, you have to
share at other times, but this is a developing field).

Where I think these postings don't do justice is future software and
communications.  I'm not super-convinced in AI (artistic illusion), but
I do think it will be funded, so communications is foremost important.
You need to have a computer say on your wrist or a GRiD Compass sized
notebook and when you start a computation (hit CR, or what ever), your
machine should hunt for available computing resources (agents), to throw
at your problem.  This idea isn't unique goes back a long time.  A
computer has to become an extension of your memory, your thinking just
as skis must become an extension of your feet while traveling over snow
(note the tips and tails of the analogy).  There's all sorts of neat
references (most recently I read this paper:  Condor -- Hunter of Idle
Workstations -- great title).  There are numerous other problems, but I
hope you can get the idea.  You need a computational companion.

I have a few other ideas, but the real ones (on paper) are internal
planning documents and hence, I can't post.  But at this moment, my IRIS
4D has a set of windows like this (for testing purposes, again, not a
brag, debugging):  Crays, an ETA-10, Ardent Titan (thru a Sequent),
Stellar (thru an Amdahl), and your usual assortment of VAXen, Convex,
Alliant, etc.  And if you think this is a lot or neat, there are places
with more and bigger.

Eugene Miya