[molpro-user] workaround: parallel M2008 stalls on startup under SuSE

[Gershom (Jan) Martin]

The following might be useful for people who run certain Linux distributions, particularly SuSE:

We had been bedeviled for a while by a strange problem with MOLPRO 2008.1: while we could run it serially just fine, parallel runs would "hang" immediately upon startup. Our cluster runs SuSE Linux Enterprise Server 10.

After a certain amount of barking up wrong trees, I installed M2008.1 on a machine in another cluster, just for the heck of it... and found that it ran just fine! Only difference between the machines: this one was running Red Hat Advanced Server 4.

Further experimentation and digging revealed that directly running  the executable without any command line arguments worked fine, and that there was a difference in the process group files: where on the Red Hat machine they would just contain the unqualified host name (say, "hartree"), on the SuSE machine they would contain the fully qualified host name (say, "hartree.weizmann.ac.il"). Apparently , the "parallel" wrapper is allergic to the latter (or something is misconfigured on our cluster). Still further digging revealed that the default /etc/csh.login of a Red Hat Advanced Server installation sets HOSTNAME to `hostname`, but that it is set to `hostname -f` under SuSE Linux Enterprise Server.

Three possible workarounds:

(a) add "setenv HOSTNAME `hostname`" to your .login
(b) if you have root privileges, modify the relevant line in /etc/csh.login
(c) edit the main "molpro" script, go to line 128, and change it from

   hostname="$HOSTNAME"

to read

   hostname=`echo $HOSTNAME | sed 's/\..*$//'`

Now we are finally able to run Molpro 2008 in parallel!

All the best,
Jan Martin