problems running on both CPUS on SMP
Josh England <[email protected]> Thu, 15 Jan 2004 11:26:18 -0800
| Newsgroups | gmane.network.myrinet.general |
|---|---|
| Message-ID | <1074194777.9800.8.camel@localhost> |
Hi,
I've got a gm2 setup on a cluster of 2-proc nodes where I can run jobs
across all nodes as long as I'm only using one of the processors of each
node. However, when I try to run any MPI job on both CPUs of a node, it
doesn't work.
The setup is:
RH7.1
gm-2.0.5
mpich-1.2.5..10
mpirun -np 2 -v ./cpi shows:
running /home/jjengla/./cpi on 2 LINUX ch_gm processors
Program binary is: /home/jjengla/./cpi
Machines file is /home/jjengla/./foo
Shared memory for intra-nodes coms is enabled.
GM receive mode used: polling.
2 processes will be spawned:
Process 0 (/home/jjengla/./cpi ) on cn58
Process 1 (/home/jjengla/./cpi ) on cn58
Open a socket on cn58...
Got a first socket opened on port 8000.
Shared memory file: /tmp/gmpi_shmem-3560212:[0-9]*.tmp
ssh cn58 cd /home/jjengla ; env GMPI_MASTER=cn58 GMPI_PORT=8000
GMPI_SHMEM=1
LD_LIBRARY_PATH=/apps/toolworks/totalview.6.3.0-0//lib:/apps/openpbs/lib:/apps/mpich/lib:/opt/gm/lib:/opt/intel/compiler70/ia32/lib:/opt/gm/lib:/opt/intel/mkl60/lib/32:/usr/totalview/lib:/opt/intel/mkl60/lib/32:/usr/totalview/lib DISPLAY=localhost:10.0 GMPI_MAGIC=3560212 GMPI_ID=0 GMPI_NP=2 GMPI_BOARD=-1 GMPI_SLAVE=10.0.0.158 /home/jjengla/./cpi
ssh cn58 -n cd /home/jjengla ; env GMPI_MASTER=cn58 GMPI_PORT=8000
GMPI_SHMEM=1
LD_LIBRARY_PATH=/apps/toolworks/totalview.6.3.0-0//lib:/apps/openpbs/lib:/apps/mpich/lib:/opt/gm/lib:/opt/intel/compiler70/ia32/lib:/opt/gm/lib:/opt/intel/mkl60/lib/32:/usr/totalview/lib:/opt/intel/mkl60/lib/32:/usr/totalview/lib DISPLAY=localhost:10.0 GMPI_MAGIC=3560212 GMPI_ID=1 GMPI_NP=2 GMPI_BOARD=-1 GMPI_SLAVE=10.0.0.158 /home/jjengla/./cpi
MPI Id 1 is using GM port 2, board 0, GM_id 3712574230.
MPI Id 0 is using GM port 4, board 0, GM_id 3712574230.
Received data from all 2 MPI processes.
Sending mapping to MPI Id 0.
Sending mapping to MPI Id 1.
Data sent to all processes.
All remote MPI processes have exited.
Reap remote processes:
ssh cn58 -n kill -9 6273 2>/dev/null
ssh cn58 -n kill -9 6274 2>/dev/null
strace shows the shmem file is being read without errors. Any ideas?
-JE