Scheduling workloads to AWS placement groups¶
Run:ai supports AWS placement groups when building and submitting a job. AWS Placement Groups are used to maximize throughput and performance of distributed training workloads.
To enable and configure this feature:
Jobs | New job.
Scheduling and lifecycleenable the
Topology aware scheduling.
Topology key, enter the label of the topology of the node.
Preferredfrom the drop down.
Required—when enabled, all PODs must be scheduled to the same placement group.
Preferred—when enabled, this is a best-effort, to place as many PODs on the same placement group.