During the upcoming scheduled downtime, we will update the default MPI module, MVAPICH, at OSC.
What is changing?
The default MPI, MVAPICH, will be updated from version 3.0 to version 4.1.
Why is this happening?
We have identified a known issue where some applications built with MVAPICH 3.0 may hang due to a Process Management Interface (PMI) mismatch with Slurm on our RHEL 9 systems. Updating to MVAPICH 4.1 resolves this issue and improves overall stability.
User Action Required:
-
New Jobs: After the downtime, the default MPI module will automatically be set to
mvapich/4.1. -
Existing Binaries: We highly recommend recompiling your applications using the new MVAPICH 4.1 toolchains to ensure compatibility and stability.
-
Legacy Support: If your workflow requires the older version, you may continue to use it by running
module load mvapich/3.0. However, please note that the hanging issue may still occur.
If you encounter any issues during this transition, please contact OSC Help or submit a request through the OSC Support Center.
