[Berzelius-users] Berzelius is back in service

[Henrik Henriksson]

Berzelius is back in operation

Dear Berzelius users,

The cluster is now back in operation, with both login nodes and compute 
available for use. However, due to some logistical issues in our supply 
chain this week, we will be operating with slightly reduced capacity, at 
least over the weekend.

The command `jobsh` is now available as an effect of the Slurm upgrade 
performed during the week. This command can be used to get an 
interactive shell on a node where you are already running a job, for 
example in order to debug or monitor an already running batch job. 
Information on how to use the command is available by running `jobsh 
-h`. As an example:

>     [hx at berzelius001 ~]$ jobsh -j 4447848
>     [hx at node053 ~]$

The possibility of using raw SSH to allocated nodes will be turned off 
shortly.

The GPUs on the login nodes are a shared resource, inteded to be used 
exclusively for visualisation. However, some users are running 
CUDA-calculations on these GPUs, and these calculations may disturb the 
users using the hardware for its intended purpose. In an attempt to deal 
with this we have disabled the possibility to run CUDA-code directly on 
the login nodes. Calculations should be scheduled on a compute node, and 
for development work we recommend using the `devel` reservation.

Please reach out to us if you have any questions regarding any of the 
above, and thank you for your patience during this downtime!


Best regards,
Henrik Henriksson
NSC