CVL Slurm cluster

The Computer Vision Lab (CVL) owns a Slurm cluster with restricted access. The information in this article is an addendum to the main Slurm article in this wiki, specific for usage of the CVL cluster. In addition to this and the main article, consult the following sources of information if the information you're looking for isn't available here:

• All articles listed under Data access

• Matrix room for update and maintenance information

• Matrix room CVL cluster community help

Access

Access to the CVL Slurm cluster is granted by Kristine Haberer.

Setting environment

The environment variable SLURM_CONF needs to be adjusted to point to the configuration of the CVL cluster:

export SLURM_CONF=/home/sladmcvl/slurm/slurm.conf

Hardware

The following tables summarizes node specific information:

Detailed information about all nodes can be seen by issuing the command

scontrol show nodes

An overview of utilization of individual node's resources can be shown with:

sinfo --Format nodehost:14,statecompact:7,cpusstate:16,cpusload:11,memory:8,allocmem:10,gres:55,gresused:62,reason:10

(Adapt the field length for gres and gresused to your needs)

Automatic/default resource assignment

• Jobs not explicitely requesting GPU resources receive the default of 1 GPU
• Jobs receive a default of 2 CPUs and 40GB per assigned or requested GPU

Limits

• Run time for interactive jobs is limited to 2 hours
• Run time for batch jobs is limited to 48 hours

Need for longer run time

If you need to run longer jobs, coordinate with your administrative contact a request to be added to the account long at ISG D-ITET support
After you've been added to long, specify this account as in the following example to run longer jobs:

sbatch --account=long job_script.sh

Display GPU availability

Information about the GPU nodes and current availability of the installed GPUs is updated every 5 minutes to the file /home/sladmcvl/smon.txt. Here are some convenient aliases to display the file with highlighting of either free GPUs or those running the current user's jobs:

alias smon_free="grep --color=always --extended-regexp 'free|$' /home/sladmcvl/smon.txt"
alias smon_mine="grep --color=always --extended-regexp '${USER}|$' /home/sladmcvl/smon.txt"

For monitoring its content the following aliases can be used:

alias watch_smon_free="watch --interval 300 --no-title --differences --color \"grep --color=always --extended-regexp 'free|$' /home/sladmcvl/smon.txt\""
alias watch_smon_mine="watch --interval 300 --no-title --differences --color \"grep --color=always --extended-regexp '${USER}|$' /home/sladmcvl/smon.txt\""

Access local scratch of diskless clients

Local /scratch disks of managed diskless clients are available on a remote host at /scratch_net/<hostname> as an automount (on demand). Typically you set up your personal directory with your username $USER on the local /scratch of the managed client you work on.

• Locally (on the client <hostname>) it is accessible under /scratch/$USER, resp. /scratch-second/$USER.
  The command hostname shows the name of your local client.

• Remotely (on a cluster node, from a Slurm job) it is accessible under /scratch_net/<hostname>/$USER, resp. /scratch_net/<hostname>_second/$USER

• On demand means: The path to a remote /scratch will appear at first access, like after issuing ls /scratch_net/<hostname> and disappear again when unused.

• Mind the difference of - used to designate a local additional disk and _ used in naming remote mounts of such additional disks

BMIC specific information

The BMIC group of CVL owns dedicated CPU and GPU resources with restricted access. These resources are grouped in partitions cpu.bmic, gpu.bmic and gpu.bmic.long.
Access to these partitions is available for members of the Slurm account bmic only. You can check your Slurm account membership with the following command:

sacctmgr show users WithAssoc Format=User%-15,DefaultAccount%-15,Account%-15 ${USER}

Notable differences

With access to the BMIC resources, the following differences to the common defaults and limits apply:

• Jobs not explicitely requesting GPU resources will not receive a default of 1 GPU but are sent to the partition dedicated for CPU-only jobs cpu.bmic

• Need for longer run time: As above, but apply to be added to bmic.long

• Run time for interactive jobs is limited to 8 hours