[diakritikal]

Not so long ago a high percentage of the Top500 were running SUSE Enterprise Linux.

I'm sure this was because at one point CRAY built one for the US DoD/DoE leveraging AMD Opertons & SUSE, and of course CRAY used to be the big name in supercomputers so I'm sure that influenced other builders choices. Then I recall the US DoD/DoE mandating a Linux OS engineered closer to home, so Red Hat got the nod.

So I wonder what the prevalence of the various distros in supercomputing is now?

[convolvatron]

just as a historical note I was involved in the OS selection for that doe machine. the customer insisted we contract to an external vendor for support, as they didn't trust us to do it in house. Suse was selected because RedHat wasn't interested in talking to us - nothing more...maybe a nod to the fact that Suse had done a fair amount of early technical work on amd64.

[lorenzhs]

My sample isn't particularly large, but RHEL and SLES still appear to be the popular choices. CentOS is also sometimes used. Maybe someone else has a better overview?

[craftyguy]

In my experience, a few systems (tens) I came across used centOS on the compute nodes, and RHEL on service/login nodes. Sometimes I would come across some custom thing (e.g. one of the big systems in China uses Kylin). But the RHEL and/or CentOS choice is by far the most popular.