errno 24 means "Too many open files".
Looks like you may be hitting the upper limit for the number of open file descriptors.
Check /proc/sys/fs/file-max and see if you need to bump it up.
Not sure if you need to bump up "ulimit -n", but worth a try.
-Aleph
On 10/14/06, Adam Moody <moody20@llnl.gov> wrote:
Hello,
I'm trying to run a 500 node job using mpirun / slurm with OpenMPI-1.1.1
and see the following errors at startup:
[rhea342:09444] [0,1,318]-[0,0,0] mca_oob_tcp_peer_recv_blocking: recv()
failed with errno=104
[rhea32:13463] mca_oob_tcp_accept: accept() failed with errno 24.
[rhea32:13463] mca_oob_tcp_accept: accept() failed with errno 24.
[rhea326:09641] [0,1,302]-[0,0,0] mca_oob_tcp_peer_recv_blocking: recv()
failed with errno=104
...
I'm starting the job with the following commands:
srun -N 500 -A
mpirun -np 500 -bynode hello_mpi
Smaller jobs around 50 nodes run just fine. Any ideas?
Thanks,
-Adam Moody
LLNL
_______________________________________________
users mailing list
users@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users