On Sun, 9 Nov 1997, Project ELLIPTIC/Wai-Mo Suen wrote:

>  Guys, 
>   
>  How are elliptic solvers working on the T3E's in Europe? 
>  In particular, Bernd, how is BAM working at present? 

Basically since last week we now have PETSc and BAM working on the T3E at 
ZIB:

- With an increased allocation, Gerd got PETSc installed, and it works
  (maximal slicing, octant and full grid)

- BAM_Elliptic (multigrid) compiled after minor changes, and it now works
  (initial data, maximal slicing, only full, only 2^n processors)

Other people have also experience with the T3E in Garching. At the moment,
it is just me doing first tests at ZIB, and "it works" means that small
test problems give the right numbers. I just started testing on more
than 65^3.

As to be expected, there still are minor technical problems when you do
something for the first time (/tmp/xx?! Thanks, Paul), and there are
surprises (why did PETSc dump core after three iterations on 129^3 in
convergence mode on Sunday? full moon?). 

In case you only heard about the problem but not that it is now solved,
with BAM_Elliptic I encountered an MPI request leak in Cactus on both the
T3E and the Origin (MPI crashed after a finite number of iterations),
which Paul finally located.

In brief, now we have PETSc and BAM_Elliptic working on the T3E at ZIB,
although the installation must still be called weakly tested.

Bernd

Created for the WUGRAV CoCoBoard Page Projects Page.

Created by The CoCoBoard.