Platform for AI (PAI) - AI scheduling engine v2.0 implements multi-level job preemption
Feb 21 2025
Platform for AI (PAI)Dynamic job preemption shares computing power and improves the utilization of resources.
Content
Target customers: Internet companies, LLM start-ups, and AI research institutions. New features/specifications: The quota-based scheduling engine of PAI uses job type classification (such as training, inference, development, and priority) and dynamic priority evaluation algorithms to trigger the preemption mechanism. This ensures that high-priority jobs can be quickly executed. In addition, combined with the AIMaster preemptive rollback technology, interrupted jobs automatically save the intermediate state and enter the queue. After resources are released, they are preferentially resumed for efficient scheduling in resource-intensive scenarios.
Help Document
https://www.alibabacloud.com/help/doc-detail/2869203.html
