ERROR; return code from pthread_create() is 11

[Kevin Kocot]

Hello,

I'm trying to run RAxML 8.2.4 (pthreads sse3 version) on our university's new HPC. I'm getting an error in my queuing software's output file (we use slurm) that says "ERROR; return code from pthread_create() is 11." I was wondering what this error code means.

Here's the code I'm using to execute RAxML with slurm:

#!/bin/bash
#SBATCH --job-name=RAxML
#SBATCH -p main
#SBATCH -n 48
#SBATCH -N 1
#SBATCH --mem-per-cpu=20G
#SBATCH -o slurm_output
#SBATCH -C R920     #this puts it on the meganode
#SBATCH -p owners   #this puts it in a higher prority partition
#SBATCH --qos kmkocot    # this says you have a right to run up to 48 cores as long as you want

export DK_ROOT=/share/apps/dotkit
. /share/apps/dotkit/bash/.dk_init
use raxml-8.2.4
for FILENAME in *.fas
do
srun --ntasks-per-node=1 raxmlHPC-PTHREADS-SSE3 -T 48 -s $FILENAME -n $FILENAME.out -f a -N autoMRE -p 12345 -x 123545 -m PROTGAMMAAUTO -q FcC_supermatrix_partition.txt
done

Any advice would be greatly appreciated.

Best,
Kevin Kocot

[Alexey Kozlov]

Hi Kevin,

it seems like you are trying to run RAxML with too many threads (see
http://stackoverflow.com/questions/7038586/return-code-from-pthread-create-is-11).

Assuming that your nodes really have 48 cores, there are a couple of things I would try:

1. The line

> #SBATCH -n 48

looks suspicious, it might be that you are starting 48 RAxML processes with 48 threads each.

2. Please make sure that your for loop is really executed sequentially (which is what you want). If 'srun' call is
asynchronous (not sure about that), you will be starting multiple RAxML instances in parallel (one per FASTA file) -
which will again result in oversubscribing the CPU cores.

Hope this helps.

Best,
Alexey

> --
> You received this message because you are subscribed to the Google Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to raxml+un...@googlegroups.com
> <mailto:raxml+un...@googlegroups.com>.
> For more options, visit https://groups.google.com/d/optout.