Oscar Upgrade (Rebuilding MPI Applications)

Oscar Users,

As part of the Oscar maintenance (8:00 am, January 5, 2021, until 11:59 pm, January 7, 2021), we are upgrading Slurm to version 20.02.6. This may require you to rebuild your MPI applications. To help facilitate this, we are providing access to a test cluster with the new version of Slurm. Please note - some of the older, existing MPI modules will be deprecated and removed from the system in the near future. Please see below for details.

1. New Cluster is for testing only

  • Intended for rebuilding and testing MPI applications
  • Has only a small number of nodes (5), so do not run computationally intensive jobs

2. Recommended new mpi modules to use in your MPI applications

  • mpi/hpcx_2.7.0_gcc_10.2_slurm20**
  • mpi/hpcx_2.7.0_intel_2020.2_slurm20**
  • mpi/openmpi_4.0.5_gcc_10.2_slurm20
  • mpi/openmpi_4.0.5_intel_2020.2_slurm20
  • mpi/mvapich2-2.3.5_gcc_10.2_slurm20
  • mpi/mvapich2-2.3.5_intel_2020.2_slurm20

** We highly recommend using HPC-X modules for new builds. From our preliminary testing, HPC-X provides significantly better performance as it leverages [SHARP protocol](https://www.mellanox.com/products/software/hpc-x) to improve the performance of MPI operations by offloading collective operations from the CPU to the switch network.

3. If your MPI applications specifically require older MPI modules, please use:

  • mpi/openmpi_2.0.3_intel
  1. Additionally, we are working to rebuild commonly used MPI applications, please feel free to test those as they become available. Please check our documentation for their build status and a new name.

5. Instructions for testing the new Slurm

  • Connect to Oscar via ssh or VNC
  • ssh login005 or login006

As always, please email support@ccv.brown.edu with any issues, questions, or concerns.

Sincerely,

CCV User Services