Earth Simulator Rapid Response Meeting May 15 , 2002

1
Earth Simulator Rapid Response MeetingMay 15 ,
2002

• Some thoughts
• Bill Feiereisen
• Los Alamos National Laboratory

2
HPC is a small top end market

• No business case??
• I believe convergence is greater than we admit
• the commercial community needs the same
  performance/architecture/capacity that we do
• But just a few years later
• However we have not made sure that our preferred
  architectures have trickled down to the latest
  commodity products

3
I believe that

• We can influence the state of processor/architectu
  re/comm/storage by supporting vendors during
  these few years
• Until these architectures prove to be a solution
  to the business plan.
• But this support must be consistent and solid
• And vendors must be able to count on it
• but they must be made accountable

4
Support during this weak part of the business
plan.

• Broad spectrum of custom vs. commodity
• Its not either/or (Beowulf ?? Custom)
• But somewhere in between
• No everywhere in between.
• Search for balance
• Not one HPC architecture but many?
• Support to vendors to explore points in this
  spectrum
• And not beat then up with our attempt to get the
  cheapest deal for each procurement
• Hold them (and us) accountable
• Its our investment in our future
• And theirs.

5
What about this GS40?

• Reaction?
• Shock at their demonstrated success
• Good on them!
• No surprise at their consistent drive to success
• Technology?
• Nothing magic, But an adaption of a commercial
  design, (albeit a low volume loss-leader?)
• This seems to be an example of govt support
  during those years
• Unanswered question Are GS40 technologies now
  making it into the commercial space?
• Will this be self sustaining? trickle down to
  smaller commercial uses?

6
Another cross agency init.?

• Been down this road before
• Question what went wrong before doing it again
• However we must do something
• Our (USA) strength is to do it together
• We have more critical mass than anyone else
• Our (USA) weakness is to do it together
• We have a short attention span
• Our ROI must come too quickly
• We react in fits and starts

7
Status?

• Not really a crisis of technology
• More a crisis of will, consistency and
  persistence
• Backed up by sufficient (considerable) resources