Hello,
I was wondering if openmpi had a -pernode like behavior similar to osc
mpiexec ....
mpiexec -pernode mpi_hello
Would launch N mpi processes on N nodes ... No more no less.
Openmpi already will try and run N*2 nodes if you don't specify -np
mpirun -np mpi_hello
Launches N*2 mpi processes on N nodes (when using torque and 2ppn
specified in your nodes file). This is good.
I tried:
$ mpirun -nooversubscribe mpi_hello
[dn172:09406] [0,0,0] ORTE_ERROR_LOG: All the slots on a given node have
been used in file rmaps_rr.c at line 116
[dn172:09406] [0,0,0] ORTE_ERROR_LOG: All the slots on a given node have
been used in file rmaps_rr.c at line 392
[dn172:09406] [0,0,0] ORTE_ERROR_LOG: All the slots on a given node have
been used in file rmgr_urm.c at line 428
[dn172:09406] mpirun: spawn failed with errno=-126
Here's our env:
$ env | grep ^OM
OMPI_MCA_btl_mvapi_ib_timeout=18
OMPI_MCA_btl_mvapi_use_eager_rdma=0
OMPI_MCA_rmaps_base_schedule_policy=node
OMPI_MCA_btl_mvapi_ib_retry_count=15
OMPI_MCA_oob_tcp_include=eth0
OMPI_MCA_mpi_keep_hostnames=1
This helps us to simply launch scripts to be generic enough to run on
1ppn and 2ppn studies pretty easily.
----
echo "Running hello world"
mpiexec -pernode mpi_hello
echo "Running hello world 2ppn"
mpiexec mpi_hello
---
Thanks,
-cdm
