Hi,

it seems unlikely that it is a problem of ADF. I've used ADF over the
last months on dual processor Athlon nodes,
without problems, and the benchmarks I've made seemed to indicate
that,
like my previous benchmarks, the time
needed scales more or less linearly with the clockspeed of the
processor, as well as more or less linearly with the
number of nodes in parallel jobs.

Therefore, how did you setup ADF for parallel computing ? Using PVM,
MPICH or LAM-MPI ?
I experienced some difficulties (read: bad performance) when using
MPICH a few years ago,
and found LAM-MPI to be both stable and fast; therefore if you have
used MPICH, this might be your problem.
[Note: I haven't tried recent MPICH versions, so it might be that it
has improved lately]

Otherwise, it seems to be related to your cluster, and not really to
ADF, I guess.

Please let me know how you setup ADF, and the parallel computing with
it,
so we could figure out exactly what might be causing this.

Best wishes,
––––––––––––––––––––––––
–––––––––––––
Marcel Swart

Organische en Anorganische Chemie
Faculteit der Exacte Wetenschappen
Vrije Universiteit Amsterdam

De Boelelaan 1083
1081 HV Amsterdam
The Netherlands

F +31-(0)20-4447488
E swart_at_chem.vu.nl
W http://go.to/m.swart
––––––––––––––––––––––––
–––––––––––––
Received on 2003-01-13 14:18:51