The current automated slurm workflow consists of 4 steps, each one in its own job and each job depending on the previous one:
sysimage
ref
sampler
extract
We currently do not support running multiple systems (eg 118_ieee, 300_ieee, 1354_pegase...) at the same time. Therefore, a user needs to run the above workflow multiple times, one for each system.
This can create conflicts for the sysimage job, since different jobs would attempt to save a sysimage at the same location (currently app/julia.so).
In addition, if a user wants to skip the sysimage step (e.g. they have already compiled one), they need to manually modify the submit.sh script in order to remove the --dependency that the ref job has on the sysimage job.
Supporting multiple systems in the same config would be more work than we can afford at the moment.
Instead, I would suggest we separate the sysimage creation from the other 3 jobs. This would add one manual step (creating the sysimage) before one can submit the ref/sampler/extract series of jobs.
The current automated slurm workflow consists of 4 steps, each one in its own job and each job depending on the previous one:
We currently do not support running multiple systems (eg
118_ieee
,300_ieee
,1354_pegase
...) at the same time. Therefore, a user needs to run the above workflow multiple times, one for each system. This can create conflicts for the sysimage job, since different jobs would attempt to save a sysimage at the same location (currentlyapp/julia.so
).In addition, if a user wants to skip the sysimage step (e.g. they have already compiled one), they need to manually modify the
submit.sh
script in order to remove the--dependency
that theref
job has on thesysimage
job.Supporting multiple systems in the same config would be more work than we can afford at the moment. Instead, I would suggest we separate the sysimage creation from the other 3 jobs. This would add one manual step (creating the sysimage) before one can submit the
ref/sampler/extract
series of jobs.