Partitions & Time Limits

Slurm schedules jobs into partitions (queues). Each partition defines policy such as default/maximum wall time, access control, hardware class, and more. Time format is D-HH:MM:SS. The * marks the default partition.

---

Example of live sinfo output (snapshot)

PARTITION   AVAIL  TIMELIMIT  NODES  STATE NODELIST
general*       up 10-00:00:0      9   idle N[01-09]
xeon128        up   infinite      2   idle N[01-02]
epyc256        up 30-00:00:0      1   idle N03
epyc512        up 30-00:00:0      6   idle N[04-09]
interactive    up 10-00:00:0      3   idle N[01-03]
DB             up   infinite      1   idle N10                # Reserved 

Snapshot table explanation

Partition	Avail	Time limit	Nodes	State	Nodelist	Notes
general*	up	10-00:00:0	9	idle	N[01-09]	Default; used if -p is not specified
xeon128	up	infinite	2	idle	N[01-02]	Nodes with 2 × Xeon CPUs, 48 cores and 128 GB RAM
epyc256	up	30-00:00:0	1	idle	N03	Node with 2 × EPYC CPUs, 96 cores and 256 GB RAM
epyc512	up	30-00:00:0	6	idle	N[04-09]	Nodes with 2 × EPYC CPUs, 96 cores and 512 GB RAM
interactive	up	10-00:00:0	3	idle	N[01-03]	Nodes enabling interactive work (srun)
DB	up	infinite	1	idle	N10	Reserved

---

How time limits work

• If you omit --time, the job gets the partition’s DefaultTime (which can be shorter than MaxTime).
• Request time explicitly with --time=DD-HH:MM:SS:

  sbatch -p general --time=2-00:00:00 myjob.sh
  srun   -p interactive --time=02:00:00 --pty bash
  

• Requests exceeding a partition’s MaxTime are rejected.
• A partition can show infinite, but site/account QoS may still cap runtime.

---

Examples

1) Submit a batch job (default partition)

# Uses 'general' (default) with its default time:
sbatch myjob.sh

# Request 2 days explicitly:
sbatch -p general --time=2-00:00:00 myjob.sh

2) Submit to a specific hardware partition

# EPYC 512 GB nodes for 12 hours:
sbatch -p epyc512 --time=12:00:00 myjob.sh

# EPYC 256 GB node for 6 hours:
sbatch -p epyc256 --time=06:00:00 myjob.sh

3) Interactive shell for debugging

# 2-hour interactive session:
srun -p interactive --time=02:00:00 --pty bash

4) Pin to specific node(s) (diagnostics only)

# Force placement on node N03 for 1 hour:
sbatch -p general -w N03 --time=01:00:00 myjob.sh

5) See partitions and nodes

# Compact summary (same columns as the snapshot):
sinfo -s

# List nodes in a partition (name, state, CPUs, memory MiB):
sinfo -p epyc512 -N -h -o "%N %t %c %m"
# Columns: NodeName State CPUs Memory(MiB)

6) Inspect a partition’s policy

# Shows DefaultTime, MaxTime, access rules, etc.:
scontrol show partition general

7) Check your jobs with partition shown

squeue -o "%10i %8u %10T %10P %N %10M %l" | column -t
# JOBID USER STATE PARTITION NODELIST TIME LIMIT