Resource sharing adaptive configuration based GPU virtualization QoS control system and method

Abstract

The invention provides a resource sharing adaptive configuration based GPU virtualization QoS control system and method. GPU resource capacity sharing and distribution among all the virtual cloud computing loads can be realized under the constraint of QoS via a monitoring module, a control module and a scheduling module. The resource sharing adaptive configuration based GPU virtualization QoS control method improves the QoS stability of concurrent operation of graphic loads in a virtual GPU resource based virtual cloud system platform remarkably, and improves load frame rate stability degree when compared with rest traditional QoS control methods. At the same time, under the precondition of maintaining stable and smooth operation of all the concurrent working loads, the GPU computing resource occupation rate is reduced obviously, and the system platform is enabled to support more working loads to keep an acceptable QoS level.

Description

technical field [0001] The technical field of virtualization of cloud system architecture of the present invention, in particular, relates to a GPU virtualization QoS control system and method based on resource sharing adaptive configuration. Background technique [0002] At present, the cloud system architecture based on virtualization technology has been widely used in various fields such as business, scientific research and education. This wide applicability in practical application fields is basically based on the inherent characteristics of the cloud platform architecture, which include high concurrency in task processing and flexibility in resource sharing. Obviously, all the above features are based on the resource scheduling, sharing and isolation functions contained in virtualization technology. However, the kinds of computing resources that can maturely apply the above functions and avoid performance degradation and hardware limitations are still very limited, and...

AI Technical Summary

Problems solved by technology

[0005] A problem in the common design of most native schedulers is that all virtual machines, whether busy or idle, take turns to become active with the same chance, without stable resource allocation or isolation

Therefore, such a system cannot cope with QoS fluctuations caused by changes in resource requirements

However, in the current system design, there is no system design that can realize the adaptive function to deal with similar problems.

Images