Thus, use the mpirun option -nolocal
Otherwise, nodes 12-15 are often used for SCIRUN and Matlab simulations. If you wish to run a job across the entire cluster, and it requires a consistent load, please send us email a day in advance, and we will schedule time for you.
If you are unsure about something, just send us email.
16+1 Dell PowerEdge 2650s (one used for head node)
There are two networks on the cluster. Both are on separate gigabit switches. As such, each node has two names: compute00-15 and generic00-15.
The basic idea is that all filesystem traffic goes over the generic switch, thus leaving the compute traffic on its own dedicated switch.
Most of this should be invisible to the user. Since all file systems are mounted across the cluster, you should not have to do any file copying between nodes. The network has been setup so that all filesystem traffic will go over the generic network by default.
So, if you need to login to one of the nodes, just use its compute name (eg, rsh compute05).
Home directories are mounted from our SAN FS server and are located
in
/home/users/$USER
Each node has a roughly 26gig disk which is viewable from all other
nodes. The mount points are:
/disks/disk-compute00
/disks/disk-compute01
/disks/disk-compute02
/disks/disk-compute03
/disks/disk-compute04
/disks/disk-compute05
/disks/disk-compute06
/disks/disk-compute07
/disks/disk-compute08
/disks/disk-compute09
/disks/disk-compute10
/disks/disk-compute11
/disks/disk-compute12
/disks/disk-compute13
/disks/disk-compute14
/disks/disk-compute15
Use these as you would like. This is more like scratch space. If you see a drive you are using fill up, move you output to a different one (or it might be deleted).