multithreads for parallel NEURON code

General issues of interest both for network and
individual cell parallelization.

Moderator: hines

Post Reply
breakwave922

multithreads for parallel NEURON code

Post by breakwave922 »

I need to do run my big-scale NEURON model some supercomputing resources (code is paralleled), but unfortunately, we are limited to get number of cores. To speedup, I was suggested to launch multiple threads by manually setting OMP_NUM_THREADS. I'm not sure NEURON can do this. My understanding is, to set OMP_NUM_THREADS, I need to use OpenMP. But the current code use OpenMPI that is not memory-share based.
Moreover, my code supports recording LFP, which uses extracellular mechanism. According to the discussion viewtopic.php?t=2004, multithreads can not be supported if extracellular mechanism is used.

But if anyone have thoughts or ideas on this, please let me know.

Thanks in advance.

Here is current batch file I used to run my job:

#!/bin/bash
#BSUB -q normal
#BSUB -n 80
#BSUB -e "stderr.%J"
#BSUB -o "stdout.%J"
#BSUB -W 3:10
#BSUB -J fmodel

module load openmpi/2.1.2-gcc_5.4.0-cuda_8.0.61
module load python/3.6.1
mpirun nrniv -mpi xx.hoc
Post Reply