You are here

ligand docking with MPI

3 posts / 0 new
Last post
ligand docking with MPI


I'm trying to use rosetta mpi to dock a library of molecules, but I got this error:

mpirun noticed that process rank 1 with PID 26757 on node c1117 exited on signal 6 (Aborted). It seems there is an error in one node and then causes the whole job aborted, so I wonder if there are any parameters to skip the molecules with problems and continue docking. Many thanks !



Post Situation: 
Wed, 2019-09-04 03:47

It depends a bit on the particular issue which causes the process to abort whether you can change it, but in general when Rosetta encounters an issue, it will stop completely rather than continuing. If you're running MPI, this means that all of the processes on MPI stop.

There are a couple of ways to mitigate this, though. The first is to pre-check all your inputs. Instead of doing a full run of all compounds (large -nstruct), you can do a short run (e.g. -nstruct 1) just to check if everything is setup correctly. This won't catch all the issues - there can be odd run-to-run issues which crop up - but it will catch the major, obvious issues, particularly the issues related to input files, or if there's a particular chemical type which Rosetta has problems with. You can then fix/remove those entries and do the full run.

Another approach is to not do things with MPI, but instead manually split the runs into multiple, single-processor groups. This way a problem with one processor won't bring down the others. (Depending on your cluster scheduler, it also is easier to schedule.) For ligand docking, each output structure is independent of every other one, so there really isn't much to gain in running under MPI, aside from avoiding the hassle of manually splitting up the jobs.

Wed, 2019-09-04 10:01

Thank you for your suggestions and that really help me a lot.

Fri, 2019-09-06 07:13