Partitions
Types of partitions/queues
In general, there are 2 types of partition in the cluster. Community partitions and priority partition. Community partitions include GPU nodes from the priority partitions. Users using the community queues may face preemption. Priority partitions contains GPU nodes that are contributed by research teams where they are given priority access.
Job preemption refers to an event where your job is stopped to free up resources for other higher priority jobs to take place. Whenever there are insufficient resources, jobs in the researchlong queue will be preempted. If your job is running on one of the compute node contributed by a research team and they require resources for their work. There will be a chance where your job will get pre-empted to free up resources.
- Student – assigned to UG/PGstudents.
- Project – assigned to UG/PG students for their project assignments.
- researchlong – assigned to researchers, max run time of 5 days
- researchshort – assigned to researchers, max run time of 2 days.
- Priority – assigned to research teams who contributed their GPU nodes to the cluster.
Quality of Service (QOS)
To ensure that resource-demanding jobs on the cluster run smoothly, QOS are used to allocate the required resources to accounts.
In general, the QOS quotas are as follows:
| QOS | CPU cores | RAM (GB) | GPU | Maximum Job Runtime (days) | Home Directory Storage (GB) |
|---|---|---|---|---|---|
| student | 4 | 16 | 1x RTX 2080 | 1 | 50 |
| project | 30 | 30 | 1x RTX 3090 | 1 | 80 |
| research | 10 | 128GB | 1x RTX 3090/V100/A40 | 5 | 80 |
| priority | Unlimited | Unlimited | Any | Unlimited | Unlimited |
If additional resources are required, special QOS can be created and assigned upon request.
For more information about your account's partition and QOS, log on to the server and execute the myinfo command.