choderalab / ensembler

Automated omics-scale protein modeling and simulation setup.
http://ensembler.readthedocs.io/
GNU General Public License v2.0
52 stars 21 forks source link

Strangeness with running build_models using mpi #49

Closed sonyahanson closed 9 years ago

sonyahanson commented 9 years ago

Made lots of meta yaml files (64 to be exact). Not sure why. Worked fine using an interactive session. Note that I was using 64 cores for 36 models, which could have contributed.

jchodera commented 9 years ago

Can you paste in the command line or script you used, and what the output filenames were?

sonyahanson commented 9 years ago

This was the runscript:

#!/bin/tcsh
#PBS -l walltime=72:00:00
#PBS -j oe
#PBS -k oe
#PBS -q batch
#PBS -l procs=64,mem=64gb,vmem=64gb,pmem=64gb
#PBS -V
#PBS -N ensembler
#PBS -M [EMAIL ADDRESS]
#PBS -m abe
cd $PBS_O_WORKDIR
mpirun -np 64 ensembler build_models
mpirun -np 64 ensembler refine_implicit

The files for which there were 64 were the build_models-meta0.yaml files.

jchodera commented 9 years ago

It sounds like all 64 processes were launched thinking they were MPI ID 0 (the master).

As mentioned in #5, this might have been due to using incompatible MPI implementations, which are hopefully sorted out now!

sonyahanson commented 9 years ago

Sorted.