You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
On ARCHER, NAMD jobs with SMP have intermittent difficulties with the ncpus and/or mpiprocs directive to "-l". This should be investigated, further. It appears to stem from the corespernode needing to be set at 1 which also sets ncpus to 1.There are several possibilities here:
A return to defaulting to not issuing ncpus or mpiprocs with the "-l" command and allowing users to specify if they are necessary or not as whether these are needed or not varies machine to machine.
Another consideration is to see if there is actually something that is missing with these NAMD jobs that would let Longbow correctly assign ncpus ie -d to aprun.
Another solution is to have corespernode and mpiprocs as seperate parameters such that mpiprocs defaults to corespernode unless mpiprocs states otherwise. This should provide a better way to calculate ncpus from corespernode and mpiprocs (most users won't need to worry about it, just those using NAMD or those that wish to undersubscribe nodes).
The text was updated successfully, but these errors were encountered:
Users can now supply "mpiprocs" in their job or host conf files to
undersubscribe without altering the corespernode parameter, which would
often cause issues.
closes#105
On ARCHER, NAMD jobs with SMP have intermittent difficulties with the ncpus and/or mpiprocs directive to "-l". This should be investigated, further. It appears to stem from the corespernode needing to be set at 1 which also sets ncpus to 1.There are several possibilities here:
A return to defaulting to not issuing ncpus or mpiprocs with the "-l" command and allowing users to specify if they are necessary or not as whether these are needed or not varies machine to machine.
Another consideration is to see if there is actually something that is missing with these NAMD jobs that would let Longbow correctly assign ncpus ie -d to aprun.
Another solution is to have corespernode and mpiprocs as seperate parameters such that mpiprocs defaults to corespernode unless mpiprocs states otherwise. This should provide a better way to calculate ncpus from corespernode and mpiprocs (most users won't need to worry about it, just those using NAMD or those that wish to undersubscribe nodes).
The text was updated successfully, but these errors were encountered: