[cam-users] about CAM2 performance on a Linux cluster

Martin Montero mmontero at tlaloc.imta.mx
Wed Feb 18 18:20:12 MST 2004


Hi everybody,
 
I am running CAM2.0.2 under a 2.4 GHz  6-dual-node Xeon Linux cluster
(compiled with PGI)
However I am afraid I am getting very low performances. Running the
control case 
with Eulerian dynamics T42L26 the highest efficiency I am getting with
all the processors (12)
is about 1 YEAR SIMULATION PER DAY, which seems very low compared to
what some benchmarck studies
are reporting with the same number of processors (~ 4 to 5 years per
day).
 
My question is to the people running under similar conditions
(architectures), if they are getting better
performances than that at all? Or maybe I am missing something on the
compilation. The model is being
compiled using the hybrid MPI/Open MP mode and it is using 1-GBit
Ethernet connections for the switch. 
Thanks in advance for any suggestions. 
 
Martin J. Montero-Martinez
Hidrometeorología, IMTA
Paseo Cuauhnáhuac 8532
Col. Progreso
CP 62550 Jiutepec, Mor.
MEXICO
===========================
Tel:  +52 (777) 329-3600 ext 824
Fax: +52 (777) 329-3683
 


More information about the cam-users mailing list