Using PVM on the kong cluster

Setup

While the version of Condor we have installed does not currently support scheduling PVM applications, running PVM in parallel with Condor works fine. It does, however, require a small amount of setup on your account so PVM can run software on the compute nodes without asking for your password.

1. Log into kong.ling.washington.edu and run ssh-keygen to create an SSH key for yourself. Accept the default file location, and use a blank passphrase.
2. Add the new key into your authorized_keys file, and restrict the file permissions:
   1. cat ~/.ssh/id_rsa.pub >>~/.ssh/authorized_keys
   2. chmod 600 ~/.ssh/authorized_keys
3. Add the host keys for the nodes to your known_hosts file: cat /opt/etc/ssh-host-keys >>~/.ssh/known_hosts

Testing

Try running pvm and adding a few of the compute nodes as hosts. You should get something like this:

user@kong:~$ pvm
pvm> add kong-n1
add kong-n1
1 successful
                    HOST     DTID
                 kong-n1    80000
pvm> add kong-n2
add kong-n2
1 successful
                    HOST     DTID
                 kong-n2    c0000
pvm> add kong-n3
add kong-n3
1 successful
                    HOST     DTID
                 kong-n3   100000

Usage notes

By default, PVM is set up to use the system's PVM installation, in /usr/share/pvm3. If the program you're running includes its own PVM implementation, you may need to change the settings of the PVM_ROOT and PVM_ARCH environment variables.

You can always find out what nodes are available by running condor_status and looking at the left column. Only the part of the hostname between the @ sign and the first dot is needed by PVM (e.g., "kong-n1").

-- DavidBrodbeck - 12 Jul 2007