Queues

When submitting jobs to the bbgcluster, with interactive or e.g., sbatch, you have the option to specify the queue/partition where sending the job. By default (if not specified), it goes to the normalpartition.

Choose the correct queue

Selecting the right queue is crucial for the proper functioning of the cluster:
- normal: Standard job in terms of time and resources. E.g, notebook, an interactive, etc
- bigmem: Job requiring a lot of memory. 512Gb at max.
- bigrun: Job requiring a lot of nodes during hours/days (a lot of cpus/mem). Normally this case is when using a (nextflow) pipeline and your main job is submitting many sub-jobs to the queue for days. Optimal for intogen/boostDM, sarek, deepUMIcaller,... jobs

List queues

Once you enter the bbgcluster, you will see in the terminal <username>@login01:

<username>@login01:~$ sinfo
PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
normal*      up   infinite      1   drng bbgn004
normal*      up   infinite      6    mix bbgn[012,014,016,019-020,022]
normal*      up   infinite      2  alloc bbgn[021,023]
normal*      up   infinite     11   idle bbgn[005-011,013,015,017-018]
bigmem       up   infinite      2    mix bbgn[020,022]
bigmem       up   infinite      2  alloc bbgn[021,023]
bigrun       up   infinite      6    mix bbgn[012,014,016,019-020,022]
bigrun       up   infinite      2  alloc bbgn[021,023]
bigrun       up   infinite      8   idle bbgn[008-011,013,015,017-018]

In here, you can see all the three partitions (normal, bigmem and bigrun) and the status of every working node (bbgnXXX). The possible states are: - drng: The node is under maintenance. - alloc: Nodes with no resources available - idle: Nodes with all the resources available. - mix: Nodes with some resources available.

Nextflow example

You can edit your nextflow config file and define the partition:

process { 
executor = 'slurm' 
queue = 'bigrun' 
errorStrategy = 'retry' 
maxRetries = 2 
} 

executor { 
queueSize = 30 
}

Reference

• Miguel Grau
• Federica Brando