Given by Scott Klasky, Tom Haupt, Geoffrey C. Fox at SIAM Workshop Minneapolis "Is Message Passing Obsolete" on 16 March 97. Foils prepared 18 March 97
Outside Index
Summary of Material
We describe basic physics and computational features of Binary Black Collision Grand Challenge |
We show a simple finite difference is complicated by
|
MPI or more generally message passing is not obsolete but rather the only complete solution! |
Higher level systems DAGH and especially HPF cannot express full complexity of problem and MPI must fill in the missing parts |
HPF2 might be able to express full problem but HPF1 misses many key capabilities |
DAGH is more succesful than HPF1 but needs MPI for parts of problem |
Postscript or FramemakerVersion |
HPF Code for Linear Waves |
Grand Challenge Alliance Home Page |
NPAC Activity in Grand Challenge Alliance |
Outside Index Summary of Material
See Original Foil |
From http://godel.ph.utexas.edu/Ftp/images/black_hole_collision_id.gif |
The gold and voilet curves are isosurfaces of the metric component gxx for field =2(gold) and 1.5 (violet). The green surface is the zero isosurface of Kxx, the extrinsic curvature, so it's like looking at the position of space which will be frozen at the initial time, since the extrinsic curvature is like the velocity. The gold balls are at the throat of the holes. These are the solution of the elliptic initial value problem. |
These are for different parts of UNIGRID codes |
HPF diff subroutine calculates differentials |
DAGH is sample solver
|
This application does NOT suggest Message Passing is Obsolete!
|
This application suggests a tool-kit approach where you can link together many environments and not be tied to a single system
|
This stresses need for interoperable subsystems rather than ambitious complete solutions that can't be completed |