Batch Scheduling

conceptstub

Gang scheduling, quotas, fair sharing, and topology-aware placement.

Planned content

• Why the default scheduler is not enough for distributed training
• Kueue: ClusterQueue, LocalQueue, cohorts, preemption, TAS
• Volcano: PodGroup, queues, gang scheduling
• JobSet and Kubeflow Trainer as job APIs
• Why HPA fails for training jobs