Sie befinden sich im Service: RWTH Compute Cluster Linux (HPC)

Current Problems

Current Problems

Detailinformation

Scheduling by Core with Hyperthreading enabled

For the time being, we have disabled hyperthreading on the compute nodes. There had been too much issues, with HT being enabled. We will look into this further.

 

Multinode Jobs, which need ssh and do not use SLURM mechanisms (like srun) to get on the remote hosts

We have installed pam_slurm_adopt in the meantime. This allows you to ssh to nodes, which belong to your running jobs. Multinode jobs with CFX (fluent maybe also) and starccm+ should be possible now.

 

X11-Forwarding

Some kind of unintuative X11-Forwarding is possible at the moment. There still does not exist something like sbatch --x11 but you can do native ssh-X11-Forwarding now that pam_slurm_adopt is installed.

You have two options for the time being:

  • use salloc to get an allocation of nodes. ssh with X-forwarding to the allocated hosts.
    • the drawback is, that salloc injects many SLURM-Variables into the environment, which still exist once the allocation ended. So please use a new shell for salloc
  • use a normal batchscript, which includes e.g. a sleep command or something like that, than ssh to the remote nodes as soon as the job runs
    • this is the preferred way and we will provide a small program to ease this for you as soon as possible
      • at the moment, we are distributing "guishell" throughout the cluster.

Drawback of that method is, that you "just" do a ssh to one compute node, so it is a plain environment without all the slurm variables, you would expect in a real job.

 

"Full" X11-Forwarding would mean, that in the batchscript you can e.g. directly start Intel Vtune or an xterm, without any additional work. Also the environment set there is the real jobenvironment. This is still to be implemented.


 Zusatzinformation

zuletzt geändert am 29.01.2021

Wie hat Ihnen dieser Inhalt geholfen?