eugene@eos.UUCP (Eugene Miya) (11/23/88)
Last week, I was given the pleasure of meeting Seymour Cray. Made a
short video tape of "What's all this about Gallium Arsenide?" [if your
school wants a professionally edited copy write:
University Video Communications
Stanford University
P.O. Box 2666
Stanford, CA 94309
Cost is $20 before Dec. 15 $35 after (VHS only). Intro by Norm Morse.
My tape went to Time.] Should be available end of January.
Anyway's he showed photos of the Cray-3 (housing only, obviously).
Had some performance graphs of the Cray-4. But aside from Seymour's
and a few other people's technical papers, I was disappointed in the
Supercomputing 88 conference (operationally SC'89 has learned from 88 and
will be better). Lots of other things, Steve Stevenson can summarize.
I think I have seen the future and it's not in supercomputing (or
"mainframes"). Oh yes there will be better machines, but I see too
little flexibility, too little short sightedness in all people. Difficulties
in parallelism are grossly underestimated in some areas.
So this note gives me a chance to post the above Stanford address,
I think the amount of fluff in comp.arch has reached the point
(partially due to the evils of cross-referencing and failure to
edit by novice net posters) and I'm 'u'ing this group. Just got too much
work to do. Remember when net.arch was incredibly quiet?
Another gross generalization from
--eugene miya, NASA Ames Research Center, eugene@aurora.arc.nasa.gov
resident cynic at the Rock of Ages Home for Retired Hackers:
"Mailers?! HA!", "If my mail does not reach you, please accept my apology."
{uunet,hplabs,ncar,decwrl,allegra,tektronix}!ames!aurora!eugene
"Send mail, avoid follow-ups. If enough, I'll summarize."aglew@mcdurb.Urbana.Gould.COM (11/26/88)
>I think I have seen the future and it's not in supercomputing (or >"mainframes"). Oh yes there will be better machines, but I see too >little flexibility, too little short sightedness in all people. Difficulties >in parallelism are grossly underestimated in some areas. > >Another gross generalization from > >--eugene miya, NASA Ames Research Center, eugene@aurora.arc.nasa.gov This is the best news I've heard of in a long time. Supercomputers and super-mainframes are one of the gravest threats to democracy in a long time. Democracy via personal-computing! Although I'm not so sure that paralellism is as difficult as Eugene says it is. Now that parallelism is available to the public, we'll see how much quicker the public can be at developing new ideas, compared to the boys in the national labs who had parallelism as their own domain for so long. This, plus the development of new algorithms, like Greenard's O(n) algorithm for the n-body problem. Greenard's algorithm in itself may spell the end of super-computing -- or a new florescence. Have you ever considered just how many supercomputing applications are versions of the n-body problem.
kwan@cad.ics.uci.edu (Andrew Kwan) (11/27/88)
In article <28200243@mcdurb> aglew@mcdurb.Urbana.Gould.COM writes: > This, plus the development of new >algorithms, like Greenard's O(n) algorithm for the n-body problem. >Greenard's algorithm in itself may spell the end of super-computing Could someone give a reference to the above mentioned algorithm? Thanks. Andrew Kwan kwan@ics.uci.edu
aglew@mcdurb.Urbana.Gould.COM (11/28/88)
The new O(n) algorithm for the n-body problem is described in An ACM Distinguished Dissertation - 1987, published by MIT Press: The Rapid Evaluation of Potential Fields in Particle Systems, Leslie F. Greengard, ISBN 0-262-07110-X. I am excited because a friend was parallelizing the previously best known algorithm for the n-body problem, Piers and Hutt's oct-tree O(n lg n) algorithm, and he was obtaining impressive speedups (it's amazing what a difference a professional programmer can make). I am excited because many of the applications supercomputers are used to investigate are, in fact, varieties of the n-body problem.
jeff@lorrie.atmos.washington.edu (Jeff L. Bowden) (11/28/88)
In article <28200245@mcdurb> aglew@mcdurb.Urbana.Gould.COM writes: <The new O(n) algorithm for the n-body problem is described in >An ACM Distinguished Dissertation - 1987, published by MIT Press: <The Rapid Evaluation of Potential Fields in Particle Systems, >Leslie F. Greengard, ISBN 0-262-07110-X. What is the n-body problem?
henry@utzoo.uucp (Henry Spencer) (11/29/88)
In article <JEFF.88Nov27225714@lorrie.atmos.washington.edu> jeff@lorrie.atmos.washington.edu (Jeff L. Bowden) writes: ><The new O(n) algorithm for the n-body problem is described in... > >What is the n-body problem? Given n objects in space, considering each as a point mass affected by no forces other than gravity (for simplicity's sake), predict their positions and velocities in the future given initial positions and velocities. For n=1, it's trivial. Newton solved n=2. Doing n=3 algebraically is incredibly hard, and common wisdom hath it that it is impossible. Not true; there is a little-known but valid solution using somewhat exotic infinite series. (It *is* provably unsolvable using "finite" mathematics.) Unfortunately the infinite-series solution is of no practical use: it converges too slowly. There are useful algebraic solutions for some very restricted special cases of n=3. For practical n=3 work, and any work for n>3, one must use iterative numerical approximations. Making them efficient for large n is hard. Hence the interest in the new approach. -- Sendmail is a bug, | Henry Spencer at U of Toronto Zoology not a feature. | uunet!attcan!utzoo!henry henry@zoo.toronto.edu
kwan@io.ics.uci.edu (Andrew Kwan) (11/29/88)
In article <JEFF.88Nov27225714@lorrie.atmos.washington.edu> jeff@lorrie.atmos.washington.edu (Jeff L. Bowden) writes: >What is the n-body problem? An example of the n-body problem would be the prediction of the movements of the planets in the solar system. There are several bodies, each of which interacts with all others by gravitational force. To simulate such a system, one might choose a time step, start with the initial position and velocity of each body, calculate the net force on each body, then the change in acceleration and velocity on each body, to finally get a change in position of each body during the time step. Then you would do this all over again. The n-body problem is a "classic" problem in classical mechanics.