Submitting a Distributed MIP as a Batch Job
Answered
Hi.
Running distributed algorithms requires distributed workers and a manager. To setup the worker nodes you need to run "grb_rs --worker" on the worker nodes.
If you are using an HPC cluster where jobs are distributed with a batch system (lsf), you can submit a distributed MIP as follows:
grbcluster batch solve DistributedMIPJobs=2 misc07.mps
How do I define master and workers in this case? Is this done through the Cluster Manager, which must be installed beforehand?
Thanks.
-
Hi Tessa,
The cluster manager and compute server are not available with an academic license. With an academic license, you can only construct a pool of workers \(\texttt{grb_rs --worker}\) which do not support the batch mode.
However, you can use \(\texttt{slurm}\) and/or \(\texttt{screen}\) to work with a batch equivalent.
Best regards,
Jaromił0 -
Hi Jaromił.
Just to make sure I have got you correctly: With an academic license, I can not submit a distributed MIP as a batch (lsf) using the batch solve command?
Kind regards
Tessa
0 -
Hi Tessa,
Correct.
Best regards,
Jaromił0 -
Hey.
One last question:
However, you can use and/or to work with a batch equivalent.
What difference would using slurm instead of lsf make?
0 -
I have not used lsf make so I cannot tell.
0 -
Ahh, ok. Basically, however, the batch solve command does not work with any job scheduling system for Linux clusters, regardless of which one it is.
0 -
Yes, that's correct.
0 -
Thank you! Before, I thought you were going to tell me that another job scheduling system would be better for this, which kind of confused me. But now it's clear.
0
Please sign in to leave a comment.
Comments
8 comments