Weblog – Thomas S. Iversen

I recently helped identifying a problem with continuing calculations from a checkpoint file performed by desmond of the Schrödinger software portfolio. The error we got was

fail to extract simulation parameters from checkpoint file <filename>

It worked for a lot of people, but not for others. I searched high and low for differences:

• Different unix environment settings.
• Differnet MPI settings (intel mpi vs. gcc mpi vs. software mpi)
• Different permission settings on home dirs
• Differences in groupmembership between those who could continue calculations and those who couldn’t.
• Performing the calculations on different nodes in the calculation clusters.

As it turned out it was neither of these points that was responsible for problems. Instead it was the use of non-english chars in the comment field in the /etc/password file for the users who could not continue the calculations from a checkpoint file. The fix for this is simple:

usermod -c "new name, only english chars" <userid>

Everything was fine after that.