>This would be great if the newest version of ADF would work with MPI on
>linux. I found that ADF2000 ran better in parallel with MPICH than with PVM
>on Alpha linux. I.e. with two CPUs (on the same node) I would see a scaling
>of 1.7 with PVM while a 1.95! scaling with MPI.
>I thought the problem with MPICH is that it doesn't support Fortran90 at the
>moment. I could be completely off base on this. Comments?
We run ADF2000 with MPI on a cluster of Pentium nodes and another cluster of
We have used both the MPICH and the LAM-MPI, where we found improved results
for the LAM implementation (faster, more stable).
I don't have exact numbers, but I remember one example, the e_QMMM_Butane job.
On 1 node, 1 task it runs on the DEC/Alpha in ca. 5 mins.
On 1 node, 2 tasks (MPICH) it hasn't finished yet at the end
of the CPU-limit of the short queue (30 mins).
On 1 node, 2 tasks (LAM-MPI) it runs also ca. 5 mins.
-- Marcel Swart Theoretische Chemie (MSC) / Moleculaire Dynamica (GBB) Rijksuniversiteit Groningen Nijenborgh 4 9747 AG Groningen The Netherlands m.swart_at_chem.rug.nl http://go.to/m.swartReceived on 2001-04-25 17:22:08
This archive was generated by hypermail 2.2.0 : 2006-11-02 07:00:02 CET