From: James Rogers (jamesr@best.com)
Date: Mon Sep 03 2001 - 13:06:37 MDT
On 8/29/01 12:53 PM, "Robert J. Bradbury" <bradbury@aeiveos.com> wrote:
> IBM has devoted the current issue of its Systems Journal
> to unveiling some of the applications it anticipates for
> "Blue Gene". In particular an article by Allen et al
> details some of the features of the architecture itself.
>
> In a paper that I hope to release sometime in September
> one of the things I mention briefly is a possible speedup
> in the path to MNT enabled by hardware like Blue Gene.
> It may be useful for the technophiles amongst the list
> to familiarize themselves with the IBM specs so when
> they read my paper they can consider whether I'm being
> excessively conservative.
I've been following the architecture and my general impression has been that
it is a huge step forward in some areas, yet is completely useless in others
such that those "others" may be rate limiting factors.
Blue Gene is somewhat asymmetric with respect to which directions it can
solve the equation given the architecture. It will be very useful, given a
problem, to do protein modeling to see what sticks (i.e. sifting). It would
seem to be far less useful for generating a solution given a particular
problem, mostly because the algorithm space looks quite a bit different in
relation to something that would fit on top of this computer architecture.
Sifting *is* a very useful capability, but it is still a blind squirrel
approach to the engineering problem. The reverse problem is arguably much
more powerful and more interesting (though much more difficult to make a
supercomputer for), at least IMHO.
-James Rogers
jamesr@best.com
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 08:10:21 MST