[scyld-users] Problem running Charmm on Scyld cluster

[Joshua Bernstein]

Hello Andre,

  I've CC'd to the list a summary of the steps required to support 
Charmm on the compute nodes. I just wanted to make sure everyone on the 
list is able to benefit.

     Below is a script you need to copy into /etc/beowulf/init.d and 
call it charmm

then:
     $ chmod +x /etc/beowulf/init.d/charmm

then reboot the compute nodes (as root)
     # bpctl -S all -R

NOTE: you may choose to reboot just one node to see if the script is 
working. Then reboot the rest of the cluster. To reboot, say node 4 do 
soemthing like:

     # bpctl -S 4 -R

run your two tests to see if the scripts worked... If it does work I 
will CC this e-mail out to the mailing list:

-----
#!/bin/bash
#
#############
# A script to pre-cache the required libraries
# for both 32 and 64-bit versions of a static charmm binary
#
# Written for Scyld cz-5
#
# By: Joshua Bernstein (jbernstein at scyld.com)
# Scyld Software
#
# Use: Put in /etc/beowulf/init.d and reboot the compute nodes. Make 
sure the script
# is set to be executable. (chmod +x charmm)
#############

[ "$NODE" == "" ] && NODE=$1
if [ "$NODE" == "" ]; then
    echo "No node specified"
    exit 1
fi

# Here for 64-bit static charmm binary support
bpcp /lib64/libnss_files.so.2 $NODE:/lib64/libnss_files.so.2
bpcp /lib64/libnss_files-2.3.2.so $NODE:/lib64/libnss_files-2.3.2.so
bpcp /lib64/libc.so.6 $NODE:/lib64/libc.so.6
bpcp /lib64/ld-linux-x86-64.so.2 $NODE:/lib64/ld-linux-x86-64.so.2
bpcp /lib64/ld-2.3.2.so $NODE:/lib64/ld-2.3.2.so
----SNIP----

In order to run the CHARMM 32-bit binary. Hopefully you have the 32-bit 
libraries installed into /opt/lib32. If you don't you'll have to use the 
CD and install them before you continue with everything below.
     After the libraries are installed you'll actually need the 32-bit 
libraries on the compute nodes. Your /etc/exports file should like like 
this:

/home  @cluster(rw)
/opt   @cluster(rw)

Notice the addition of /opt.

Now as root run:

# exportfs -a
# /etc/init.d/nfs restart

Now you'll need to edit /etc/beowulf/fstab to force the compute nodes to 
mount /opt in the proper location. The end of my file looks like this:

$MASTER:/home   /home           nfs     nolock,nonfatal         0 0
$MASTER:/opt    /opt            nfs     nolock,nonfatal         0 0

Again notice the added line for /opt. The next two commands will reboot 
the compute nodes, and restart the Beowulf service. The two commands 
must be typed in quick succesion in order to avoid having to manually 
power cycle the nodes in the machine room:

# bpctl -S all -R
# /etc/init.d/beowulf restart

Now, the 32-bit CHARMM should work as expected...

-Joshua Bernstein
Software Engineer
Scyld Software