Difference between revisions of "Rolling Upgrade from RHEL7 to RHEL8"
Moskalenko (talk | contribs) |
Moskalenko (talk | contribs) |
||
| Line 27: | Line 27: | ||
We recommend using the following environment modules to rebuild your code for EL8: | We recommend using the following environment modules to rebuild your code for EL8: | ||
| − | * gcc/12.2. | + | * gcc/12.2.2 openmpi/4.1.5 |
| − | * cuda/12.2. | + | * cuda/12.2.2 gcc/12.2.0 openmpi/4.1.5 |
* intel/2020 openmpi/4.1.5 | * intel/2020 openmpi/4.1.5 | ||
| − | * cuda/12.2. | + | * cuda/12.2.2 intel/2020 openmpi/4.1.5 |
| − | * nvhpc/23. | + | * nvhpc/23.7 openmpi/4.1.5 |
==Known Issues== | ==Known Issues== | ||
Revision as of 20:46, 13 September 2023
HiPerGator OS Upgrade: RHEL-7 → RHEL-8
HiPerGator is beginning rolling upgrade process from Red Hat Enterprise Linux version 7 (RHEL-7) to version 8 (RHEL-8). The upgraded software environment will include updates to important components including, but not limited to, NVIDIA GPU drivers, CUDA (12.x), communication libraries, compilers (gcc/12 and intel/2023), and OpenMPI (4.1.5).
Some programs/workflows will run in the updated software environment without any modification, and some will require recompilation or adaptation.
What will happen
On August 16, 2023, a subset of HiPerGator compute resources will go online with the updated software environment. On the same day, a new login node pool will be introduced as hpg-el8.rc.ufl.edu that runs the updated software environment. If you need to rebuild your application, you can do it here.
Over the next 6-7 weeks, additional compute resources will be upgraded at a steady rate until they are all upgraded. Once more than half of the compute resources are upgraded, the new login nodes will become default as hpg.rc.ufl.edu.
We recommend moving to RHEL-8 at your earliest convenience.
Choosing RHEL-7 or RHEL-8
Using the '--constraint' SLURM directive, it is possible to submit jobs specifically targeting RHEL-7 with '--constraint=el7' or RHEL-8 with '--constraint=el8'. If neither is specified, the scheduler will not consider operating system in placing your jobs and you may get either or a mix depending on the number of requested nodes.
#SBATCH --constraint=el8
or
#SBATCH --constraint=el7
How long will RHEL-7 systems remain available on HiPerGator?
At this time, we expect to complete the upgrade process on September 20, 2023. If you have concerns about access to RHEL-7 systems on HiPerGator after September 20, 2023, please contact us as soon as possible to discuss.
Rebuilding Code for EL8
We recommend using the following environment modules to rebuild your code for EL8:
- gcc/12.2.2 openmpi/4.1.5
- cuda/12.2.2 gcc/12.2.0 openmpi/4.1.5
- intel/2020 openmpi/4.1.5
- cuda/12.2.2 intel/2020 openmpi/4.1.5
- nvhpc/23.7 openmpi/4.1.5
Known Issues
Missing PMIX_V2
If you see an error similar to the following
srun: error: Couldn't find the specified plugin name for mpi/pmix_v2 looking at all files
or
error while loading shared libraries: libpmix.so.2: cannot open shared object file: No such file or directory
Then, your job likely landed on en el8 node, which doesn't have pmix v2. Make sure to rebuild your code against a modern OpenMPI stack (openmpi/4.1.5) to work on EL8 before the end of the transition. Until the code is rebuilt please use the el7 SLURM constraint to continue running your analyses. Note that the number of el7 nodes is dwindling and is going to reach zero by September 20th. Rebuild your code now!