[CentOS-devel] What should I do in my cluster? CentOS streams dilemma.

Fri Dec 25 13:24:49 UTC 2020
redbaronbrowser <redbaronbrowser at protonmail.com>

‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Friday, December 25, 2020 4:54 AM, Pablo Sanz Mercado <pablo.sanz at uam.es> wrote:

>     I have been reading a lot about the new CentOS stream
> "revolution", but now I am trying to understand the dilemma most of
> "regular" sysadmin have:
> What should I do?
> As you know there are a lot of computational centers working
> with CentOS. Sometimes because it is the best OS sysadmin knows,
> sometimes because of a low budget problem.
> Now we have this important change, so we have to decide. We
> have to choose the OS we are going to install in our servers, moreover
> we have to decide as quickly as we can, because as you know changes
> should be done as calm as posible.
> Could we list pros/cons about using CentOS stream in a
> computational Center?
> Are we going to see CentOS stream in the top500 list?
> Best regards,
> Pablo Sanz

To start off for a computational cluster running in-house written applications, I would have three primary/storage nodes running RHEL 8.

All of the other nodes would be diskless.  I would boot Fedora (one version previous to latest) via PXE/NFS and then non-OS application/data would be on a gluster mount.

I want my focus for high availability and administration to be on just three servers.  The rest of the servers should be easy to spin up/down as well as upgrade/replace as needed.

Pro of using CentOS Stream would be you get features before they are released for RHEL.

Con of using CentOS Stream is it won't get offcial support and packages will be released before getting the same level of QA of RHEL 8 or CentOS 8.  The other con is CentOS Stream will not be getting major software updates/features as frequently/soon as Fedora.

It is /possible/ you may see CentOS Stream appear on the top 500.  I think it is unlikely.  To make it on the bottom of the list you need over 20,000 cores.  With a budget that can get you that much equipment and the resulting on-going costs, paying for RHEL and getting the support probably will seem worth it.

The majority of my nodes wouldn't be running RHEL to save money.  It is just Fedora would provides me the kernel and library versions I would prefer to work with on the computation nodes.  I would also rather self-support the OS for those nodes.

If there is a specific application you are building the cluster for then you may want to ask the software vendor what OS/distribution they recommend using.