|
High Performance Computing
|
|
August 28, 2008
|
Assigning Hosts Using LSF
|
MPI Documentation
|
|
HP Superdome Cluster |
The load-sharing facility (LSF) allocates one or more hosts to run an
MPI job. In general, LSF improves resource utilization
for MPI jobs that run in multihost environments. LSF handles the
job scheduling and the allocation of the necessary hosts and HP MPI
handles the task of starting up the application's processes on the hosts
selected by LSF.
By default mpirun starts the MPI processes on the hosts
specified by the user, in effect handling the direct mapping of host names to
IP addresses. This is a problem because when jobs are submitted to LSF,
the user generally does not select the hosts his job will run on - LSF
selects them based on the cluster load when the job is run, which may
be quite different from the situation when the job was submitted.
To work around this problem a script called runpam has been provided;
if you want to examine it, the full pathname of the file is
/opt/cluster/bin/runpam. It gets information on which hosts to use from
LSF, then creates an internal appfile by which it uses with
mpirun to run the program. More information is available about this
process under the
MPI section in the FAQ
, and in the
MPI Documentation section
of this webpage.
If you want to run a mpi job which has sub-processes using a parallelized
library routine, or a routine parallelized using OpenMP or other loop
parallelization techniques, follow this link.
Send comments/questions to:
help-hpc@uky.edu
Last modified: November 17 2003 13:39:35.