After help from members of this list, I was able to get lam-mpi functioning
under LSF on hpux 11.23 (the workaround was to unset the LSB_JOBID and
LSB_JOBINDEX environment variables). When I try to scale up from the
neighborhood of mpi processes, I start getting problems similar to the ones
I had previously - lamboot runs, but using mpirun to run the software
produces a message that the lamd process isn't present. lamd is running -
at least, lamnodes returns correct information after lamboot is run.
Are there any other environment variables that have to be unset, or similar
kludges needed to get 6.5.9 to work?
I am appending a copy of script used to run the job. The actual mpi program
to be run, and its arguments, are arguments to the script.
Thanks
Chuck Fisher
chuck_at_[hidden]
------------------------------------------------------------------------
#!/usr/bin/csh
set tokens=($LSB_MCPU_HOSTS)
set user=`whoami`
if (-e lam-bhost.def) then
echo "lam-bhost.def already exists in" `pwd`
echo "exiting"
exit
endif
@ num=0
while ($#tokens)
echo "$tokens[1] cpu=$tokens[2]" >> lam-bhost.def
@ num = $num + $tokens[2];
shift tokens
shift tokens
end
set opath=( $path )
set path=( /scratch/vviti/AeroSoft/lam-6.5/bin $path )
setenv LAMHOME "/scratch/vviti/AeroSoft/lam-6.5"
setenv LAMHELPFILE "/scratch/vviti/AeroSoft/lam-6.5/etc/lam-helpfile"
setenv AEROSOFT_HOME "/scratch/vviti/AeroSoft"
setenv AEROSOFT_ELMHOST "@sdx0.uky.edu"
unsetenv LSB_JOBID
unsetenv LSB_JOBINDEX
lamboot -x lam-bhost.def
mpirun -np $num $argv
lamhalt
rm lam-bhost.def
set path=( $opath )
|
