CRI Gardner upgrade news


Haky Im


November 30, 2020

cri gardner upgrade

  • Operating System Upgrade - The operating system will be upgraded from Red Hat Linux 6.7 to 7.6. This will provide a kernel that will allow for a more modern software ecosystem. For example, software such as tensorflow will not run on Red Hat 6.

  • GPFS Upgrade - GPFS storage clients will be upgrade from GPFS 4.2 to GPFS 5. This will provide a performance increase for metadata operations such as creating, listing, and deleting files.

  • SLURM Scheduling - The Torque/Moab scheduling on the system will be replaced with SLURM. SLURM is an open-source scheduler that has become a de facto standard across many HPC sites.

  • Deep Learning capabilities - Last year, the CRI purchased two deep learning servers with 8 NVidia V100s per server. These servers have been open to users who requested the capabilites. With the upgrade, the deep learning systems will be added to the general scheduling queue.

  • Increased container capabilities - With the upgrade to Red Hat 7, we will be able to provide the ability for users to create their own singularity containers.

  • Authentication/Authorization - We are running authentication/authorization clients that are rather outdated at the moment. Upgrading those clients should provide a more stable environment when logging into the cluster and accessing files.

  • Upgraded compilers - The compilers on the cluster will be upgraded to the latest version. This will be gcc-10.2.0, llvm-11.0, intel-2020.2, and nvhpc-20.9. The compilers will provide implementation of the latest standards for C, C++, and Fortran.

  • Enhanced accounting - We will be provided accounting on both jobs submitted to the cluster as well as software use across the environment.