GPU resource pool scheduling system and method

Abstract

The invention discloses a GPU resource pool scheduling system, and the system comprises a GPU cloud computing power center and a GPU cloud control node; the GPU cloud computing power center comprisesa plurality of GPU computing power units, and each GPU computing power unit comprises a VMM and an RC, wherein the GPU cloud control node comprises an RS; the GPU computing power units are used for providing computing power of GPUs; each VMM is used for providing a control interface, receiving a resource scheduling instruction sent by the RS, creating a virtual machine according to the instruction, allocating vGPU resources to the virtual machine and starting the virtual machine; each RC is used for counting resource data of the GPU computing power unit and reporting the resource data to the RS; each RS is used for collecting resource data reported by each RC and sending a resource scheduling instruction to each VMM, and scheduling resources of the GPU computing power unit globally, including the steps of gathering GPU resources to form a plurality of groups of GPU hardware sets, and forming a GPU resource pool by the plurality of groups of GPU hardware sets. According to the system, unified resource pool scheduling management of various manufacturers, GPU models and GPU virtualization modes in a cloud computing platform is realized; the invention further discloses a GPU resource pool scheduling method.

Description

technical field [0001] The invention relates to the technical field of cloud computing resource management, in particular to a GPU resource pool scheduling system and method. Background technique [0002] With the generalization of GPU, the use of GPU in data centers is increasing day by day, and it has very important uses in the fields of graphics and image rendering, parallel computing, and artificial intelligence; the important elements of data centers are mainly computing, storage and network; with the cloud With the in-depth development of computing, how to manage GPU resources more effectively in the data center has become very important. [0003] The various components of cloud computing and the computing part usually need to use virtualization technology. Virtualization technology includes CPU and GPU virtualization technology. The details of physical hardware are shielded through virtualized hardware, so that physical resources can be used more flexibly; while compu...

AI Technical Summary

Problems solved by technology

[0008] Although a variety of GPU manufacturers, product lines, and models can be used for different usage scenarios, such as image rendering and scientific computing; however, due to the existence of various GPU manufacturers, product lines, and models, it is difficult to implement the current GPU virtualization technology. Moreover, for large-scale GPU computing power centers, the business form is very complex, and there are usually multiple application scenarios such as graphics rendering, scientific computing, artificial intelligence, and oblique photography modeling; in actual business use, there are a large number of graphics in the GIS industry. Rendering and virtual desktop requirements, as well as a large number of oblique photographic modeling requirements and artificial intelligence requirements, while the user's graphics workstations, GPU servers and other resources are independent of each other, becoming information islands, unable to share resources, and the conversion of GPU resource usage scenarios Re-deployment of physical servers usually takes several days; therefore, how to effectively manage and schedule these GPU resources in the cloud computing architecture becomes particularly important

Images