GPU dynamic scheduling method and device based on real-time load and electronic equipment

Abstract

The invention relates to the technical field of graphics processors, and discloses a GPU dynamic scheduling method and device based on real-time load and electronic equipment, the GPU dynamic scheduling method comprises the following steps: obtaining service alignment index data, and calculating derivative index data reflecting service stability and reliability; and guiding allocation scheduling of resources and services based on the alignment index data and / or the derivative index data. According to the GPU dynamic scheduling method based on the real-time load, for a cloud GPU use scene, scheduling planning is carried out through real-time GPU load information, the bottom layer GPU resource difference is shielded, the use efficiency of GPU resources is fully explored, and the scheduling complexity is reduced.

Description

technical field [0001] The present invention relates to the technical field of graphic processors, in particular to a real-time load-based GPU dynamic scheduling method, device and electronic equipment. Background technique [0002] In actual scenarios, many GPU-based services cannot fully utilize a GPU graphics card, which will cause a waste of graphics card resources. There are many types of GPU graphics cards. Different graphics cards have different load capacities and memory specifications. The same service deployment structure may not work well on another GPU model. [0003] Regarding the scheduling of GPU services, the following technical methods currently exist: [0004] 1. Manually mixed deployment scenarios. [0005] 2. In the cloud scheduling scenario, define the GPU memory usage of the service itself in the k8s service orchestration, and the scheduler checks whether the memory is free for stacking or tiling. [0006] Problems with current technology: [0007] ...

AI Technical Summary

Problems solved by technology

[0007] 1. Manual mixed deployment scenarios not only consume a lot of manpower and material resources, but also often rely on the accumulation of experience, which is difficult to handle for large-scale GPU service scenarios

[0008] 2. The cloud scheduling scenario does not take into account the real load capacity of GPU services. Using stacking processing cannot meet the service requirements of small video memory and high load. High-load service stacking will lead to a decline in service quality; using tile processing will be subject to service scheduling. Sequence, and waste of graphics card resources

[0009] In actual use, GPU usage scenarios, due to the framework selection of the service itself, code implementation, and the diversity of card models, there will be a lot of waste of GPU computing power; the existing cloud environment does not have mature real-time information based on the actual use of GPU resources. Scheduling plan

Images