Software

Problems with MVAPICH2

The default MPI installation on Oakley and Ruby, mvapich2/2.1, appears to have a bug that is triggered by certain programs.  The symptoms are 1) the program hangs or 2) the program fails with an error related to Allreduce.

To test whether a failure is related to this issue, as opposed to an error in your code, set the following environment variable in your batch job:  MV2_USE_SHMEM_COLL=0  This option disables optimizations.  If your program runs correctly, you can blame the MPI library.

There are several workarounds to choose from.

Glenn module lammps-7Dec15 bug

Batch scripts loading module lammps-7Dec15 should use the user's login shell or

the Korn shell, e.g. #PBS -S /bin/ksh

There is a bug that causes the module load to fail if the job script specifies the Bash shell, i.e.: 

#PBS -S /bin/bash

Alternatiely you can unload mpi and intel-10.x before loading the lammps module.

An example failure is:

Issue when loading multiple Fluent or ANSYS modules simultaneously

Due to the way our Fluent and ANSYS modules are configured, simultaneously loading multiple of either module will cause a cryptic error.  The most common case of this happening is when multiple of a users jobs are started at the same time and all load the module at once.  In order for this error to manifest, the modules have to be loaded at precisely the same time; a rare occurrence, but a probable occurrence over the long term.  

 

If you encounter this error you are not at fault.  Please resubmit the failed job(s).

Certain modules not accessible

Certain modules are not working for all clusters since the downtime.  We have reports specifically that Amber, Gaussian, and Turbomole are not working.  We are working to resolve the issue, but until then jobs using Amber, Gaussian, Turbomole, or other software with permissions restrictions will not work.

Subscribe to RSS - Software