Resource sharing method and device

Abstract

A resource sharing method is applied to a Kubernetes cluster, and comprises the step of starting an MPS Server Pod corresponding to a GPU on any node of the Kubernetes cluster according to the sharinguse requirement of a container or the Pod for the GPU of the node. The invention further provides a resource sharing device. According to the method and the device, the MPS Server Pod can be dynamically started in the Kubernetes cluster, and the GPU can be shared and used by the Kubernetes container or the Pod, so that the use efficiency of cluster resources is improved.

Description

technical field [0001] The embodiments of the present application relate to but are not limited to the field of computer application technology, especially a resource sharing method and device. Background technique [0002] Due to the powerful computing power of the Graphics Processing Unit (GPU), it is widely used in the fields of deep learning and high-performance computing, but the requirements for using the GPU vary in different scenarios. Taking deep learning as an example, in the training scenario, multiple GPUs need to be aggregated to provide greater computing power to accelerate model training; however, in the inference process, it is more desirable to share the same GPU among multiple applications, so as to Share the excess computing power when a single application uses the GPU. At the same time, with the popularity of container technology, more and more applications use containers and container clouds as tools for orchestrating and scheduling applications and for...

AI Technical Summary

Problems solved by technology

However, starting MPS directly through static mode will limit the use of GPU to shared

In this way, when there is no job using the shared GPU, it will lead to long-term occupation of GPU resources, resulting in waste of resources; moreover, it is impossible to adaptively manage the shared or exclusive use of the GPU

Images