daemon@utcsri.UUCP (The devil himself) (12/25/85)
Conference Review: (97 lines)
------------------
First International Conference on Supercomputing Systems
- SCS 85 -
Tampa, Florida December 16-20, 1985
Sponsored by IEEE Computer Society
1) Altho billed as "international", the US was the main focus of this
conference. This was sometimes embarrassing during the plenary
sessions when speakers urged the need for the US to regain its
leadership in computer technology.
2) Supercomputing? Many of the papers had nothing to do with
supercomputing! It seems that most of the papers were invited
and the organizers were hard put to find enough papers on the
subject!
3) Each day was encumbered by morning plenary sessions where leaders
of (mainly US) technology and politics attempted to say something
prophetic about supercomputing and high tech. During the week,
four speakers each gave an almost identical summary of the history
of "supercomputing". One speaker gave a state of the art summary
of current architectures that so superficial it was silly!
Most of the comments made by the VIP Captains of Industry
were vacuous, and many of the more serious issues of
funding, training, availability of supercomputers for research,
etc. were overlooked. There was no time for questions from the
floor. Dropout rate after the second day was quite noticeable.
4) Like so many conferences these days, at this one the (almost)
complete proceedings were available at registration. This practice
tends to make actual attendance at the conference somewhat
optional, because you can now read the papers at your leisure
rather than sit thru a probably boring recital. A better idea
would be to either make the proceedings available a few weeks
before the conference and allow more time for questions/comments,
or to delay the printing until weeks after the conference to
encourage participation in the sessions.
As it turned out, there were some many papers crammed into the
parallel sessions that there was usually no time for significant
discussion and session moderators had their hands full maintaining
time discipline. Also, only 10 or 15 minutes was available to
give each paper, so what in actuality was presented was a mere
abstract of the paper.
5) At the Grand Banquet, over an hour was spent in an awards ceremony
which was as ridiculous as it was unnecessary. Everyone on the
plenary panels received (dubious) awards for their participation.
So did all the conference organizers. This left everyone wondering
who decided the awards in the first place. It looked like the
organizers were giving themselves awards! The crowd did start
getting restless and some of the tables were having a hard time
keeping it together. There was alot to be sarcastic about!
(I almost passed out from laughter on an empty stomach during
this Emperor's-new-clothes cerimony!)
6) By the last day of the conference, about 30% of the original
attendees showed up. This indicates that it was too long (without
the plenary sessions, it could have been over with in 2.5 days
rather than 4! This would have been a significant savings for
my employer! One would have hoped to get a lot more out of a
four day conference!
7) The technical sessions were clearly hardware oriented. There
were no papers or sessions on operating systems or compiler
technology for supercomputers! I believe these are the most
significant problems for the industry right now, even more
significant than architecture and chip design! As far as I'm
concerned, this oversight was tragic!!
8) Some useful information was obtained in spite of it all:
a) The NEC SX-2 is twice as fast as the CRAY X-MP (single
processor ). It has a 6 ns clock and 40 vector registers of 256
elements each. And its (FORTRAN) compiler is top class.
b) IBM's vector processor 3090 is a serious entry. The 3090/200
with 2 processors, and the 3090/400 with four, have 18.5 ns clock
with measured megaflop speeds of 60 to 90 for real problems.
It has a cached memory and this has caused them to redo the
algorithms for FFT and LINPACK to accomodate this.
9) Overall, I'd give this conference a C+ grade. There was too much
splash and glitter and not enough substance. I'd like to propose
that someone organize a conference on compiler technology for
super/parallel computers. Perhaps a workshop rather than a
conference would be more appropriate. And it should be soon...
Very soon! It is really needed.
----------------------------------------------------------------------
Richard Friedman: Pacific - Sierra Research Corp.,
2855 Telegraph Ave., Suite 415, Berkeley CA 94705
(415) 540 5216
uucp: {hplabs,lll-crg,ptsfa}!well!rchrd