Hey guys~~
Great work on the 7.0 stuff -- I can't seem to break it...yet.
One thing I am seeing is that mpirun will fail to find a file if the NFS
mountpoints do not match on the cluster. We use automout to mount
master:/usr/home to /mnt/home on the node. In both places /home is a
symlink to that home directory. When a user tries to run a program on
the nodes, it complains:
bash-2.05$ more lam-debug-log.txt
started (7.0b12cvs), uid 27659, gid 27659
kernel: initialized
Link 0: node: 0, cpus: 1, type: 388, ip: 192.168.2.5
Link 1: node: 1, cpus: 2, type: 0, ip: 192.168.2.10
Link 2: node: 2, cpus: 2, type: 0, ip: 192.168.2.17
Link 3: node: 3, cpus: 2, type: 0, ip: 192.168.2.16
Link 4: node: 4, cpus: 2, type: 0, ip: 192.168.2.11
Link 5: node: 5, cpus: 2, type: 0, ip: 192.168.2.14
Link 6: node: 6, cpus: 2, type: 0, ip: 192.168.2.13
Link 7: node: 7, cpus: 2, type: 0, ip: 192.168.2.12
flatd: flqload - successfully created file
/tmp/lam-henken_at_node0.internal/lam-flatd0
flatd: flqload - file descriptor 15
flatd: flqload - successfully appended 4 bytes to
/tmp/lam-henken_at_node0.internal/lam-flatd0
kenyad: pqcreating with rtf 0x469210
kenyad: checking for directory /usr/home/henken
kenyad: looking for executable "cpi" in directory "/"
kenyad: could not find cpi, errno 2
kenyad: badness - can't execute cpi
Is there someway to get LAM to not lookup the hard path to the file,
rather the current dir + filename ?
Nic
--
Nicholas Henke
Penguin Herder & Linux Cluster System Programmer
Liniac Project - Univ. of Pennsylvania
|
