[Bi-users] Bi reboots (was: [NSC-users] Reboot of NSC login nodes and possible performance impact to new jobs)

[Kent Engström]

Mats Kronberg <kronberg at nsc.liu.se> writes:
> Dear NSC users,
>
> This afternoon we will perform an urgent security update of our
> clusters Triolith, Gamma, Bi and Elvis.

The Bi login nodes have now been rebooted with the update.

Compute nodes are being handled by a "rolling node upgrade" as usual, so
running batch jobs should not be affected.

> The update is critical and in our opinion cannot be avoided or postponed.
>
> Unfortunately, the update has a performance impact. A slowdown of
> anywhere between 0 to 30% has been reported for various synthetic
> benchmarks. However, NSC's estimate is that most HPC applications will
> not be significantly affected.
>
> If you find that jobs started after this afternoon run significantly
> slower, you can:
>
> - Extend the time limit of already running jobs (using the
> "nsc-boost-timelimit" command, or on systems where that is not
> available, contact NSC Support (https://www.nsc.liu.se/support/).
>
> - Increase the time limit of new jobs before you submit them (the "-t"
> option to sbatch).
>
> If you find HPC applications or job types that are significantly
> affected by this update, we would like to hear about it. In this case,
> email your findings to support at nsc.liu.se.
>
>
> All login nodes will be restarted as soon as we're ready to perform
> the update. You can expect 10 minutes advance warning via a message
> ("wall") to logged-in users.
>
> All compute nodes will be updated as soon as the current jobs ends, so
> all new jobs will start on updated nodes.
>
>
> You can find more information about the vulnerability at e.g
> https://meltdownattack.com/ and
> https://access.redhat.com/security/vulnerabilities/speculativeexecution

-- 
Kent Engström, National Supercomputer Centre
kent at nsc.liu.se, +46 13 28 4444