Am Montag, den 27.02.2006, 09:27 -0700 schrieb Tim S. Woodall:
> Hello Emanuel,
> You might want to try an actual hard limit, say 8GB, rather than
> unlimited. I've run into issues w/ unlimited in the past.
Oh, stupid me! On Debian /etc/pam.d/rsh does not contain
session required pam_limits.so
After adding this line it worked. I was sure that the limits were set
mpirun -np 4 bash -c ulimit -l
always returned unlimited, but
rsh node02 ulimit -l
just showed 32 which is indeed way too small. The additional bash call
seems to change the limits and fooled me this way.
Just another question: Is is possible to use a 32 Bit machine as head
node (without being involved into the computations)? When I try to run a
program using mpirun from the 32 Bit machine I get
[headnode:32560] [0,0,0] mca_oob_tcp_recv_handler: \
invalid message type: 0
and the debug output shows
[node01:01330] [0,0,1]-[0,0,0] mca_oob_tcp_peer_recv_blocking: \
recv() failed with errno=104
which means "Connection reset by peer". In between the 64 Bit machines
it works fine now.