From: Neale, Christopher Andrew (cneale_at_lanl.gov)
Date: Tue Apr 12 2022 - 10:52:15 CDT

Hello,

I had an issue with VMD and didn’t realize that my posts were getting discarded for being over the 40,000 character limit (mostly due to VMD log cut/pastes). For posterity, in case somebody else runs into a similar issue, here is a shortened version of my original email that I wanted to post at the outset:

I find an error message when I run VMD under srun and allocate fewer cores to VMD than are available on the node.

### Here are two commands that work fine:

$ /usr/projects/mrmdesign/exe/CHICOMA/VMD/vmd-1.9.4a55/exec/bin/vmd

$ srun -n 1 -c 256 --exclusive /usr/projects/mrmdesign/exe/CHICOMA/VMD/vmd-1.9.4a55/exec/bin/vmd

### Here is a command that does generate output image, but throws a bunch of error/warning messages saying: “wkf_thread_set_self_cpuaffinitylist: sched_setaffinity: Invalid argument”

$ srun -n 1 -c 1 --exclusive /usr/projects/mrmdesign/exe/CHICOMA/VMD/vmd-1.9.4a55/exec/bin/vmd

### Here’s John Stone’s advice, which worked great:

Regarding your question about the CPU affinity errors:

  These errors are occuring because your job scheduling system is preventing

VMD from modifying CPU affinity mappings.

VMD will continue to run fine despite being unable to pin particular threads

to particular CPUs, it just won't necessarily run quite as fast as it would

if it had complete control over these thread affinity mappings.

You can tell VMD how many CPU threads to use (rather than matching the

number of CPU cores times the SMT depth), by using these two environment

  VMDFORCECPUCOUNT

  WKFFORCECPUCOUNT

And for Tachyon with:

  RTFORCECPUCOUNT

### Specifically, when I added the following lines to my test, everything worked without complaint:

export VMDFORCECPUCOUNT=1

export WKFFORCECPUCOUNT=1

export RTFORCECPUCOUNT=1

Thank you,

Chris.