News from RCIC and elsewhere

HPC3/HPC2 Downtime June 15, 2022 and Aug TBD, 20202s

Thursday, May 19 2022 by Philip Papadopoulos

We periodically need to perform maintenance on HPC3 where all users are logged out and all jobs are stopped. We have two planned outages for this summer. The highlights of upcoming changes addressed by these two outages include:

  • DUO two-factor authentication will become standard on HPC2/3 login

  • Parallel File System and ZFS updated to latest stable releases

  • CentOS 7 will be sunset and Enterprise Linux 8 (EL8, Rocky Linux) will be the new OS The entire application software stack will be rebuilt for EL8. Older versions will be retained where possible, new application versions will be added

  • The version of Slurm will be updated to the latest stable release

While we normally prefer to have only a single downtime, the changes are large enough that we will handle the changes in two distinct phases.

Phase 1 ( June 15, 2022, All day)

  • BeeGFS and ZFS file system updates

  • Turn on Two-Factor (Duo) authentication

Phase 2 ( Aug TBD, 2022, All day)

  • Reinstall all nodes with EL8

  • Reinstall all rebuilt applications

  • Update Slurm

The first downtime (June 15) should only have the apparent effect on users that Duo authentication will now be required for password-based login to hpc3 and hpc3.

The second downtime (Aug) will be much more impactful In general, any user-compiled code will need to be rebuilt. Some conda environments may also need to be rebuilt to work properly with new system libraries. A few older applications simply will not build on EL8 Some widely-used versions of software - e.g. R version 3, older versions of R 4 are not buildable with all R modules under EL8. As the summer progresses, we will keep a list of sunset software.

Is keeping CentOS7 an option? Not really. It will be end-of-life in Mid-2024. We are already seeing commercial vendors ending support for CentOS7. The viability of CentOS7 as a functional OS will progressively diminish over the next 24 months.