GPU资源调度方法、装置及可读存储介质
By dynamically adjusting the distribution of GPU node labels and combining global historical data and node idle factors, the fragmentation problem of GPU resource scheduling in the cloud platform is solved, achieving balanced allocation and efficient utilization of resources.
CN116126508BActive Publication Date: 2026-07-17CHINA MOBILE COMM LTD RES INST +1
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA MOBILE COMM LTD RES INST
- Filing Date
- 2021-11-12
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
In cloud platforms, GPU resource scheduling suffers from fragmentation, leading to overuse and wear and tear on some nodes, resulting in resource waste. Existing scheduling strategies cannot effectively address this issue.
Method used
By dynamically adjusting the distribution of GPU node labels, utilizing global GPU usage history data and resource quota data, and combining node idle factors, priority is given to scheduling to idle nodes to avoid overuse and achieve balanced resource allocation.
Benefits of technology
It solves the GPU fragmentation problem, avoids the loss caused by excessive node use, and achieves efficient utilization and balanced allocation of resources.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116126508B_ABST
Abstract
本发明提供一种GPU资源调度方法、装置及可读存储介质。该方法包括：接收GPU资源请求；根据所述GPU资源请求，执行第一操作；其中，所述第一操作包括以下一项或多项：以GPU节点标签的取值作为亲和性级别，将GPU资源请求优先调度到所述GPU节点标签的取值与GPU资源请求数相同的节点上；以GPU节点标签的取值作为亲和性级别，将GPU资源请求以次优先级调度到所述GPU节点标签的取值相比于所述GPU资源请求数较小的节点上；以及以GPU节点标签的取值作为亲和性级别，将GPU资源请求以反亲和调度到所述GPU节点标签的取值相比于所述GPU资源请求数较大的节点上，在执行了所述第一操作后，基于全局GPU使用历史数据和 / 或全局GPU资源配额数据，动态地调整GPU节点标签的分布。
Need to check novelty before this filing date? Find Prior Art