Hardware : Dual-Opteron 246 Motherboard Tyan S2885
Network : Intel Gigabit NIC  + 3COM switch
OS : Fedora Core 2 i386 + OSCAR 4.0 + SIS 3.5.2 (Source)
Cluster : 4 Nodes + 1 Front-end = 5 nodes with OSCAR
 
Hi,
I think i've some trouble with switcher and my environment set up.
Root, I built a new Lam-7.0-ifort MPI soft with intel compiler from source and install intel compiler and Lam-7.0.-ifort in /opt on my front-end and nodes
I edit a new module file said "lam-oscar-7.0-ifort" which include paths to Lam-7.0-ifort and apply change with the command "switcher mpi --add-name lam-oscar-7.0-ifort <my/dir/path>".
User, I run "switcher mpi = lam-oscar-7.0-ifort" and if i type "switcher mpi" or echo $PATH, result are okay, my MPI environnement is modified accordingly. If i reboot the cluster, change are not lost.
 
If i made some verification on my nodes, by ssh and look the file "switcher.ini" inside $My-Front-end:/opt/env-switcher, the default MPI is unchanged. How to propagate change cross the cluster ?. Is this really necessary, because my Home is NFS mounted ?
 
Very strange, when I try to run a simple program "mpihello.f" with qsub My_PBS_script, I have some bad issue in the error/output file "mpihello.out", like "command lamboot not found", "mpirun not found", "lamhalt not found". If i run the job without pbs, i have echo from my nodes.
 
What could be wrong ? Many thanks
 
Best regard,
Jerome Lef�vre
-------------------------------------
Laboratoire d'Oc�anographie Physique
IRD  - Noum�a - Nouvelle-Cal�donie

Reply via email to