GPU resource management and intelligent scheduling method for deep learning

Abstract

The invention discloses a GPU resource management and intelligent scheduling method for deep learning, and the method comprises the following steps: 1, enabling a user to submit deep learning jobs through a front-end interface module, wherein the deep learning jobs comprise a to-be-executed deep learning program and a training data set; 2, after verification, adding the job to a queue to be scheduled corresponding to the scheduler; 3, starting an independent job manager for the job; 4, applying for computing resources required by job operation from a resource manager; 5, carrying out feature modeling and analysis on the to-be-scheduled job; 6, generating a resource scheduling scheme according to the operation characteristics and the cluster computing node characteristics; 7, scheduling thejob to a specified computing node according to a scheduling scheme; 8, enabling the job executor to start the container and execute a deep learning program. According to the method, the problems of low GPU resource utilization rate and poor job execution performance of an existing cluster resource scheduling method in a deep learning scene can be solved.

Description

technical field [0001] The invention relates to the technical field of cluster resource scheduling, in particular to a deep learning-oriented GPU resource management and intelligent scheduling method. Background technique [0002] Research and practice in recent years have shown that compared with traditional machine learning techniques, deep learning can achieve higher accuracy in fields such as computer vision and speech recognition, and thus has been widely used. There is a large amount of calculation in the training process of the deep learning model, and the Graphics Processing Unit (GPU) can perform this simple but large-scale computing task more efficiently, so it has become an important basic computing resource for running deep learning programs. [0003] Since GPU cards are usually expensive, it is costly to deploy an independent private cluster for each user (group), and users do not always perform model training, so users usually share these GPU resources to reduc...

AI Technical Summary

Problems solved by technology

[0008] Purpose of the invention: In view of the problems and deficiencies in the prior art above, the purpose of the present invention is to provide a deep learning-oriented GPU resource management and intelligent scheduling method to solve the problem of low utilization of GPU resources in the existing system in deep learning scenarios, Problems with poor job execution performance

Images