KU Community Cluster Operating System Upgrade
Date of Upgrade
Starting at 8:00 AM CST on January 15, 2025 and ending on 8:00 AM CST January 17, 2025, CRC will be upgrading the OS on the KU Community Cluster from CentOS 7 to Rocky 9.
This upgrade will involve updating the Operating System as well as the software packages made available to users. Previously compiled software will most likely not work on the new OS, and will need to be recompiled.
Minor or no changes
- All user files will remain unchanged
- Scheduler, partitions, time limits will stay the same
- Changing node names for easier management on our part. This only effects you if you request a specific node by name or exclude a specific node by name which we don't recommend
- The majority of software modules previously available will remain available, sometimes with version changes
Signficant changes
- Anaconda introduced licensing restrictions to their conda repo.
- We have migrated our conda to miniforge which uses the conda-forge repo by default.
- We recommend you reinstall any conda environments you have
- The path
/panfs/pfs.local
will no longer be available. Replace with/kuhpc
. - Significant changes to system software have occurred, the Standard Environment modules are now:
- compiler/gcc/11.4 replaces compiler/gcc/8.3
- openmpi/5.0 replaces openmpi/4.0
- If you have software you compiled yourself, it will likely not not run on the new system, and will need to be recompiled
- PGI compiler is now NVC (compiler/nvc/24). NVIDIA merged them in 2020
- AMD CPUs and GPUs mixed into sixhour. Need to specify
intel
/amd
ornvidia
/amdgpu
constraint if you wish to run on specific hardware - Older C6320 which have 24 cores will not have Infiniband hooked up. The ConnectX-3 card (56 Gb/s) is not supported by latest NVIDIA IB software.
Test Cluster
We have set up a test cluster to be used before January 15 to test your code and recompile any software you may have. Owners have very graciously allowed us to borrow a variety of nodes with different hardware to have a good mix of nodes to test your jobs on.
You may run your processing on the login nodes if you desire. The 60 minute CPU time limit has been removed for the test cluster.
Access
All current cluster users may access the test cluster at any time. You can use any of the current connecting methods to connect to the test cluster as well.
Login Nodes
- login1.hpc.crc.ku.edu
- login2.hpc.crc.ku.edu
Hardware and Features
Features are requested under the --constraint
option. Because the cluster is consortium of hardware, attributes allow the user to specify which type of node they wish to use (e.g. ib, edr_ib
, intel
)
Node | Cores | Memory GB | CPU | GPU/SSDs | Features |
---|---|---|---|---|---|
mi226 | 48 | 192 | AMD EPYC 9454P 48-Core Processor | 3 x MI210 64GB | avx512_only,avx2,avx512,hdr100_ib,ib,amd,mi210,amdgpu |
cre24 | 28 | 256 | Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz | avx2_only,avx2,noib,10gb,intel | |
sjm21 | 24 | 96 | Intel(R) Xeon(R) Gold 6226 CPU @ 2.70GHz | avx512_only,avx2,avx512,intel,noib | |
cho01 | 48 | 384 | Intel(R) Xeon(R) Gold 6240R CPU @ 2.40GHz | avx512_only,avx2,avx512,intel,edr_ib,ib | |
sho00 | 40 | 192 | Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz | 2 x V100 16GB 1 x A100 40GB | avx512_only,avx2,avx512,intel,edr_ib,ib,a100,v100,double,nvidia |
jac00 | 32 | 128 | Intel(R) Xeon(R) Gold 6326 CPU @ 2.90GHz | avx512_only,avx2,avx512,ib,intel | |
bio12 | 48 | 128 | Intel(R) Xeon(R) Gold 6342 CPU @ 2.80GHz | 4 x A40 48GB | avx512_only,avx2,avx512,intel,noib,a40,single,nvidia |
big55 | 56 | 256 | Intel(R) Xeon(R) Gold 6348 CPU @ 2.60GHz | avx512_only,avx2,avx512,intel,hdr100_ib,ib | |
tho00 | 48 | 256 | Intel(R) Xeon(R) Gold 6442Y | avx512_only,avx2,avx512,intel,hdr100_ib,ib |
Partitions
There are 2 partitions set up for testing. The 30min
partition has higher priority than sixhour
, meaning jobs, if submitted at the same time, the 30min job will start first.
Partition | Time Limit | Nodes | Restrictions |
---|---|---|---|
30min | 30 minutes | All | N/A |
sixhour | 6 hours | All | Limited to 100 cores per job |
Software
We installed the most used software from the KU Community Cluster. If you do not see the software that you need, please send a software request and make sure to say in the additional information section it is for the Rocky 9.
Anaconda License Restrictions
While Anaconda does state that the imposed licensing restrictions do not effect accredited universities, we are still taking precautions with this OS upgrade to move away from using the Anaconda channel. We suggest all users re-create their conda environments using the new conda/latest
module which uses the conda-forge channel.
You should not notice any change in using the conda/latest
module from the old one.
conda
, the command, is the same- You can still use the
bioconda
channel - Your previous conda environments will still work with the new miniforge install, but we do recommend to recreate them