|Cluster||Full Resource Allocation||Fair Share||Closed Access|
Full Resource Allocation means that jobs will only be run if they have a valid job 'account code' or 'project code' with a positive balance of available hours. See our Resource Allocation pages for details on how to apply for and how to use the project code.
Fair Share means that the more you use the cluster, the lower your relative priority will become. This allows new users and/or users who have not used much time recently to go higher up the queue, and allows a `fair' usage of the cluster.
Closed Access means that only certain groups have access to the system. See our Access Policies for more.
General Queuing Configurations
Each cluster has two main queues/partitions:
debugpartition, with a small number of nodes and a short wall-time. This is to be used for small test or development runs. The short wall-time enables a very quick turn-around, and faster access.
computepartition, which comprises the bulk of the cluster. The wall-time will be a number of days (between 2 and 4, depending on the specific cluster). Most jobs should be submitted here.
Please refer to '
sinfo' on each cluster for more specific limits and available resources. See this page for more on Slurm commands.
- There is a full resource allocation policy in place for this cluster.
- Jobs are scheduled based on a fair share policy as well as allocated hours.
- The maximum job size that a user should submit on this cluster is 128 cores (1 Infiniband switch worth of machines).
- Larger job sizes are now supported, but will have a performance cost, due to spanning of switches.
- There is a reservation called '
application' for interactive applications and for applications with larger memory footprints.
- This is no longer a partition, as was previously configured. See this page for details on how to use it.
- The '
application' reservation is restricted for certain projects during business hours (8am-8pm Mon-Fri), to allow for interactive usage. If unsure, please contact Ops.
- These nodes have more RAM than regular nodes (32GB or 64GB, instead of the regular 16GB nodes).