GPU (Graphics Processing Unit) resource scheduling method and system based on multi-dimensional combination parallelism

Abstract

The invention discloses a GPU (Graphics Processing Unit) resource scheduling method and system based on multi-dimensional combination parallelism. The method comprises the following steps of: 1, adding a GPU resource plug-in module in a GPU resource management center; 2, running a plug-in module, and completing corresponding linkage of various types of GPU resources of a data layer and a physical layer; 3, the GPU resource management center collects real-time basic information of GPU resources from the GPU heterogeneous cluster and issues the real-time basic information to a GPU resource scheduling module; 4, when the GPU resources are called, a request task supporting multi-dimensional combined scheduling is sent to a GPU resource scheduling module; and 5, the GPU resource scheduling module converts the main task of the request task into an executable single-dimensional scheduling sub-task and issues the sub-task to the GPU heterogeneous cluster, and the GPU heterogeneous cluster allocates GPU resources as required according to the sub-task. According to the invention, the utilization rate of GPU resources can be effectively improved.

Description

technical field [0001] The invention relates to the fields of graphics processors and computing technologies, and in particular to a GPU resource scheduling method and system based on multi-dimensional combined parallelism. Background technique [0002] In the era of AI (English full name Aritificial Intelligence, Chinese translation for artificial intelligence), algorithm engineers need to perform a large number of deep learning tasks, usually using Docker containers as training environments, using expensive Graphics Processing Unit (Graphics Processing Unit, referred to as GPU) cards can be Significantly improve the training speed, and distributed training is also the most frequent training scenario at the moment. When algorithm engineers need to use GPU resources, they need to allocate GPU resources. On the one hand, how to maximize the use of idle GPU resources is a problem that needs to be solved; on the other hand, the performance of different types and models of GPU c...

AI Technical Summary

Problems solved by technology

First of all, for the problem of resource utilization, the traditional solution is generally based on the optimization of the cloud-native cluster scheduler, and the transformation cost is relatively high; and for the problem of exclusive resource sharing, each cloud platform generally adopts node locking and resource pooling solutions. On the one hand, reasonable resource allocation needs to be carried out in advance; on the other hand, the static pooling of resources has the disadvantage of insufficient utilization

Images