[Krypton-users] Krypton service stop on November 1

[Kent Engström]

Dear Krypton Users,

a reminder about the service stop on Krypton tomorrow.

As stated earlier, we will have to cancel the jobs that are in the SLURM
queue when the service stop begins (as we need to be able to test the
queueing system "as users"). If jobs are submitted via cron during the
service stop, we reserve the right to kill them too.

We will also block logins to the login node (and log out users)
during most of the service stop. Please use Gimle if you need access to
/home and the Accumulus file systems.

As we let you in again we will update you about the changes to the
queueing system, but we aim at you being able to submit normal
jobs (i.e. non-risk jobs on normal, non-fat nodes) without changes
to the sbatch/interactive command lines.

kent at nsc.liu.se (Kent Engström) writes:
> following discussions between NSC and our SMHI user representatives
> Anders Höglund and Michael Kolax we are now ready to update the
> scheduler configuration on Krypton from the initial simple
> one to a one with more features, such as:
>
> *) risk jobs (jobs that can be preempted by non-risk jobs)
> *) simple access to the fat/huge nodes
> *) node limits between SMHI user groups
>
> Have jobs running or queued during this work would make realistic
> testing of the new configuration harder. Therefore, we are announcing a
> service stop on
>
>     Thursday November 1, from 10:00
>
> We reserve the whole working day, but hope we will be done as early as
> possible. We will also use this slot for normal system software
> updating.
>
> We will get back to you with information on how you submit different kind
> of jobs after the change.
>
> Please note:
>
> *) Use a good "-t TIMELIMIT" to sbatch/interactive during the coming
>    week if you want your jobs to be able to run before the service stop.
>    If you submit max length 7 day jobs after 10:00 today, they will not
>    start as the job will not be able to finish in time.
>
> *) Jobs that are left in the queue when the service stop starts will be
>    cancelled and have to be resubmitted afterwards.
>
>
> Regards,

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