Skip to content

Partitions

In Slurm, partitions define job execution environments by grouping compute nodes with similar characteristics and applying scheduling policies to them.

Partitions typically specify:

  • Hardware characteristics (e.g., CPU nodes, GPU nodes, memory size)
  • Job limitations (e.g., maximum runtime, number of nodes)
  • Scheduling parameters (e.g., priority tiers or quality of service)

Each job must be assigned to a partition either explicitly by the user or implicitly through a default partition configured by the cluster administrators.

The exact partitions available depend on the cluster configuration. Refer to the Devana partitions and Perun partitions for the list of available partitions and their limits.

To select a partition with a Slurm command, use the -p option:

srun | sbatch | salloc -p <partition> [...]

Example:

sbatch -p gpu myjob.sh

This submits the job to the partition named gpu.


Viewing Partitions and Their Definitions

Existing partions, included nodes, and general system state can be determined by the sinfo command.

Existing partitions and their current state can be displayed using the sinfo command.

sinfo output

sinfo
  PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
  cpu          up 1-00:00:00     10   idle n[001-010]
  gpu          up 2-00:00:00      4    mix n[011-014]
  testing      up      30:00      2   idle login[01-02]

The columns show:

  • PARTITION – partition name
  • AVAIL – whether the partition is available
  • TIMELIMIT – maximum allowed runtime
  • NODES – number of nodes currently in this state
  • STATE – node state (idle, alloc, mix, etc.)
  • NODELIST – nodes belonging to the partition

The * next to a partition name indicates the default partition.

Partitions may appear on multiple lines because nodes can be in different states simultaneously.


Detailed Partition Information

More detailed partition configuration can be displayed with:

scontrol show partition

or for a specific partition:

scontrol show partition <partition>

Displaying information for a partition

scontrol show partition gpu

This command shows configuration parameters such as:

  • maximum runtime
  • node limits
  • allowed accounts
  • memory limits
  • scheduling priority factors

Choosing the Right Partition

Selecting the appropriate partition is important for efficient scheduling. Jobs submitted to partitions with shorter time limits or more available resources are often scheduled faster.

If your job requirements exceed the limits of the available partitions, contact the HPC support team via the helpdesk.

Created by: Andrej Sec