Lend and borrow resources across teams with AWS Batch’s quota management for SageMaker Training jobs
AWS Batch quota management is a new scheduling capability you can use to divide GPU capacity into quota shares per team or project, automatically lend idle capacity across shares, and use preemption to run your highest-priority SageMaker Training jobs first, all without manual coordination.
AWS Batch introduces quota management to help organizations share GPU resources across teams and projects when training machine learning models. By defining quota shares, administrators can allocate and share constrained GPU capacity using configurable resource-sharing strategies. These strategies include reserving idle capacity, lending capacity to other teams, or a combination of both.
A quota share operates as a virtual queue nested within a job queue, allowing AWS Batch to prioritize quota shares under capacity limits and perform cross-share preemption when necessary. Additionally, administrators can enable priority-based in-share preemption to ensure higher-priority jobs run sooner. To implement quota management, a Scheduling Policy containing a quotaSharePolicy, a Service Environment specifying capacity limits, and a SageMaker Training job queue are created.
Quota shares can be associated with the job queue, and a sample notebook demonstrates how to configure these resources using the SageMaker Python SDK.
Written by urgent.news from AWS Machine Learning's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
