Resource scheduling device, resource scheduling system and resource scheduling method

Abstract

The invention discloses a resource scheduling device, a resource scheduling system and a resource scheduling method. The resource scheduling device comprises a GPU (Graphics Processing Unit) adaptivemodule, a GPU discovery module and a GPU mounting module, wherein the GPU adaptive module is used for detecting the number of GPUs of a device to which the GPU adaptive module belongs, and a driving file of the GPUs, creating a GPU calling file and informing of the GPU discovery module; the GPU discovery module is used for receiving a notification of the GPU adaptive module, monitoring usage information of the GPUs of the device to which the GPU discovery module belongs, and reporting the number of GPUs of the device to which the GPU discovery module belongs and the usage information to a management node; and the GPU mounting module is used for receiving a container-establishing request to establish a container, selecting the GPU to which the GPU mounting module belongs, and mounting the driving file and the calling file of the selected GPU into the container. According to the resource scheduling device, the resource scheduling system and the resource scheduling method, based on a containerization technique, through self-detection, monitoring, selection and mounting of the GPUs at each node, the utilization efficiency of the GPUs is improved, and the condition of non-uniform usageof the GPU resource in a cluster is avoided.

Description

technical field [0001] The present invention relates to the technical field of cloud computing resource management and container deployment, and in particular to a resource scheduling device, a resource scheduling system, and a resource scheduling method. Background technique [0002] Cloud computing is a computing mode of resource sharing, which provides available, convenient, and on-demand network access, and the resources provided by cloud computing can be quickly provided and shared and reused. Graphics Processing Unit (GPU) is a special resource in high-performance cloud computing systems. In the fields of big data and image recognition, it is necessary to use GPU to analyze and calculate data, especially for image recognition. The processing speed is much higher than that of CPU. [0003] However, the current mainstream technology is still to provide a computing environment by creating a virtual machine to mount a GPU. The application with the publication number "CN10...

AI Technical Summary

Problems solved by technology

But this will cause two problems: one is that for the underlying virtual machine, it cannot know the optimal GPU node in the entire cluster; the other is that it consumes computing time in the process of continuous upward transmission

However, in the current patents using containers, for example, CN106453492A "container scheduling method under the Docker container cloud platform based on fuzzy pattern recognition", CN105893010A "host scheduling method and system for container deployment", CN106027643A "a kubernetes-based The resource scheduling method of the container cluster management system" and so on, have not realized the automatic discovery, automatic adaptation and mounting of the GPU, and have not added the calculation weight of the GPU resources, so the GPU resources cannot be used in the container cluster

Images