Gromacs with SLURM

User discussions
1

GROMACS version: 2020
GROMACS modification: No

Good morning,

I’m running gromacs on a computing cluster using SLURM, but the simulation requires more than the job timelimits allowed (24 hours). How can I manage long simulations without submitting continuously new jobs?

Any advice would be greatly appreciated.

Mattia

2

Hi,

Not sure what you mean when you say “without submitting continuously new jobs”, since you will have to submit more. Here’s a few notes on that:

For the mdrun command, make sure to include the options -cpi -maxh 24 along with your other options. -cpi will resume from the last checkpoint, -maxh 24 ensures that mdrun stops and finishes writing all files before the 24 hours are up.

For SLURM, you can specify that your new job depends on a previous job to finish. This is how you can queue up multiple jobs to execute after each other, without running simultaneously. The documentation lists the various options to specify this. I usually submit with “singleton”, which uses the submitted job name as a check – if a job with the same name is already running, it’ll wait for that to finish. So I’ll just repeat my sbatch command to submit my job script a few times, which creates the queue since the name will be the same for all.

You can set the dependency in the job script with the SLURM_JOB_DEPENDENCY variable or use it with sbatch.

Regards,
Petter

3

Thanks a lot, you’ve been very helpful.

Regards,
Mattia