Current State

7 Day History

Notes

In many cases multiple partitions draw from the same pool of nodes, and MSI balances each partition to ensure that jobs from one partition don't hog all of the nodes in the shared pool. It is not possible to determine how many nodes are busy and available in the individual partitions.:

• msismall: Nodes in the msismall partition are shared with the msilarge and msilong partitions
• msigup: The msigpu partition has ~7 nodes with h100 gpus that are only in the msigpu partition (as shown here), but the a100 and v100 nodes are also members of the msigpu partition

• Idle: Entire node is available to run a job
• Running: Some but not all resources (memory and CPUs) on the node are allocated to one or more jobs
• Busy: All resources (memory and/or CPUs) on the node are allocated to one or more jobs
• Down: The node is not running a job and is not available to run a job

Note that the scheduler may not assign a job to an idle node, if nodes are reserved for an upcoming maintenance day or idle nodes are being accumulated for a large job, for example. The down state indicates the node has crashed, is offline for maintenance, or is not available for other reasons.