[molpro-user] Molpro 2006.1 on cluster with infiniband

Jyh-Shyong Ho c00jsh00 at nchc.org.tw
Mon Jan 8 03:05:45 GMT 2007


Dear Kirk,

Thanks, but it does not work, I get the same error message.

Jyh-Shyong Ho

Kirk Peterson wrote:

> Hi,
>
> try something like:
>
> /opt/chem/molpro/bin/molpro -n$NCPU -d$JOBDIR --mpirun-machinefile \ 
> $PBS_NODEFILE   $JOB.com
>
>
> -Kirk
>
>
> On Jan 7, 2007, at 6:01 PM, Jyh-Shyong Ho wrote:
>
>> Hi,
>>
>> I  got an Infiniband /OpenIB version of GA library and built  
>> Molpro2006.1
>> on our Opteron cluster.  However, I don't know how to set the
>> WRAPPER-mpi varibale correctly in order to run parallel Molpro job
>> under TORQUE environment.  Perhaps someone can provide me some help?
>>
>> Normall, the command to run an MPI job over Infiniband interconnect  
>> on our
>> cluster is
>>
>> /opt/mvapich/pathscale/bin/mpirun_rsh -ssh -hostfile $PBS_NODEFILE   
>> -np $nproc $executable
>>
>> Where $PBS_NODEFILE is a temporary file which include the names of  
>> nodes assigned
>> by TORQUE for this job.  $nproc is the number of processors required.
>>
>> I defined WRAPPER_mpi in file CONFIG as
>> WRAPPER_mpi="/opt/mvapich/pathscale/bin/mpirun_rsh \-ssh"
>> since both $PBS_NODEFILE and $nproc variables cannot be defined during
>> the compilation.
>> I am not able to run the compiled Molpro2006.1 with  command in a  
>> TORQUE script file
>>
>> /opt/chem/molpro/bin/molpro -n$NCPU -d$JOBDIR $JOB.com
>>
>> I always got an error message:
>>
>> Cannot find MPIRUN machine file for machine ch_gen2 and  architecture 
>> LINUX .
>> (No device specified)
>>
>> Which means that the machine list in file $PBS_NODEFILE  was not read.
>>
>> Any suggestion on how to solve this problem?
>>
>> Jyh-Shyong Ho, Ph.D.
>> Research Scientist
>> National Center for High Performance Computing
>> Hsinchu, Taiwan, ROC
>>
>>
>





More information about the Molpro-user mailing list