5001.PiMaster My-OpenMP-Scrip* pi 0 Q dev
5001.PiMaster My-OpenMP-Scrip* pi 0 Q dev
5001.PiMaster My-OpenMP-Scrip* pi 0 Q dev
5001.PiMaster My-OpenMP-Scrip* pi 0 Q dev
Sorry, I changed the pbsnodefile, here is the result:
SSH is enabled and the default password for the āpiā user has not been changed.
This is a security risk - please login as the āpiā user and type āpasswdā to se$
/var/spool/pbs/aux/5004.PiMaster
pi01
pi01
pi01
pi01
pi02
pi02
pi02
pi02
In previous PBS work, I used to do:
#PBS -l nodes=n030:ppn=2
to assign the job to specific node. It is not working with the current version, maybe you can advise the valid expression to try.
Thanks for being patient.
#PBS -l select=1:ncpus=2:host=n030
Please try the above
Thank you
Ok, that works in assigning the node, what is the syntax if I want to assign the job to node1:2processors and node 2:3processors?
Now I understand that the $PBS_NODEFILE is where the nodes and processors are assigned, so that file will need to be the -hostfile (or machinefile) in the mpiexec command. Do we need to still tell mpiexec -np ānumber of nodesā?
Thanks and by the way, some papers may end up being published and we would like to acknowledge your contribution, please email the way you want that to be cited.
Best
#PBS -l select=1:ncpus=2:host=node1+1:ncpus=3:mpiprocs=3:host=node2
#PBS -l select=1:ncpus=2:mpiprocs=2:host=node1+1:ncpus=3:mpiprocs=3:host=node2
Please try one of the above
If the MPI library is compiled with PBS TM , then it is not required. If it is not compiled with PBS TM then you have to specify
mpiexec -np < number of processes to use > -hostfile $PBS_NODEFILE
Thank you, pleasure is mine
Thank you Adarsh, can you please advise how to set MPI Library to compile with PBS TM?
Best
For example: OpenMPI
Please check this section: 10.8 Integration Using the TM Interface
In this document: PBS Professional 2020.1 Administratorās Guide , AG-446
Link: https://www.altair.com/pdfs/pbsworks/PBSAdminGuide2020.1.pdf
Hope this helps
Works nice, both of suggested format for the nodes are good. What if the line is too long for assigned nodes and PPN? How about $PBS_NP and $PBS_NUM_PPN? They donāt seem to output anything.
Will try to run a lengthy job and test the broadcasting. Is there a script that shows the performance of the broadcasting in parallel programing? I found a latency test but only between 2 nodes.
Thank you again for your great help.
Best
The line can be as long as possible, we have customer requesting for 1000 chunks.
Please node nodes and PPN are not part of openPBS syntax, however they work keeping the backward compatibility. Use select and ncpus for requesting resources
Please note this is openPBS (formerly PBS Pro OSS) . These variables seems to be from torque
work load manager.
Please check the man pages for qsub : man qsub
Please check the documentation from these links:
https://www.altair.com/pdfs/pbsworks/PBSUserGuide2020.1.pdf
https://www.altair.com/pdfs/pbsworks/PBSAdminGuide2020.1.pdf
I am not sure there isnāt a readymade script for this requirement, it is more involved and there might be many papers published on this using against the high speed network technologies and MPIās, which can be a good reference comparison & understanding.
Thank you and well noted, will do some reading of the documents.
Stay safe
I had the same problem, solved
Install all packages when installing Linux
It seems to happen when there are no server related packages.