Limits enforced by the Slurm Scheduler
The table below lists the partitions (queues) that are configured in our slurm scheduler,
along with the applicable limits for those partitions. Please note the partition for the GPU nodes is not active, pending the upgrade of the operating system on those nodes to CentOS 6. The data below is current as of April 25, 2017.
| Partition Name |
Time Limit (DD-HH:MM:SS) |
MaxTRES |
MaxTRESPU |
MaxJobsPU |
MinTRES |
| dawson |
8-08:00:00 |
cpu=512 |
|
30 |
cpu=5 |
| ellis |
8-08:00:00 |
cpu=4,node=1 |
cpu=80 |
45 |
|
fenx |
30-00:00:0 |
cpu=40,node=16 |
cpu=40,node=16 |
|
|
fielder |
30-00:00:0 |
cpu=512,node=96 |
cpu=96,node=12 |
|
|
| gque |
8-08:00:00 |
cpu=32,node=1 |
|
|
|
| jassby |
30-00:00:0 |
cpu=96,node=6 |
cpu=96,node=6 |
|
|
| kruskal |
8-08:00:00 |
cpu=512 |
cpu=512,node=32 |
8 |
|
| mccune |
3-08:00:00 |
cpu=256 |
cpu=256 |
|
|
| mque |
8-08:00:00 |
cpu=64,node=1 |
|
2 |
|
| sque |
8-08:00:00 |
cpu=512,node=100 |
|
|
|
Notes:
- Helpful Slurm Terminology:
-
TRES = Trackable RESource. Resources consumed by jobs that Slurm can track: CPUs, nodes, RAM, etc.
-
PU = Per User
-
MaxTRES = The maximum amount of various trackable resources that can be used by a single job.
-
MaxTRESPU = The maximum amount of the various trackable resources that can be used by a single user across all their running jobs.
-
In this context, cpu = core.
-
If a job exceeds MaxTRES, it will stay in a pending state, shown as 'PD' in the output of the squeue command. The REASON column, which is the last column in the output of squeue will giv the reason why the job is in the pending state.
-
The per-user (PU) limits apply to jobs running, not queued up. Once enough jobs are submitted that your PU limites are reached, any additional job submissions will be accepted, but will remain queued until enough running jobs have completed so that a pending job can start running without exceeding the per-user limits.
- To submit jobs to the Ganesh cluster, submit them to the 'mque' partition. The 'm' in 'mque' stands for 'memory', and the Ganesh nodes have larger amounts of RAM than the other nodes.