General purpose computation virtualization implementation method based on dynamic library interception

Abstract

The invention discloses a general purpose computation virtualization implementation method based on dynamic library interception. For overcoming the defects of an existing virtual machine technology in supporting general purpose computation, all functions relevant to the general purpose computation in a virtual environment are called and redirected to a non-virtual environment by intercepting and redirecting a general purpose computation runtime library in real time. Because the non-virtual environment has the general purpose computation capacity, the actual general purpose computation tasks will be completed and the results will be returned to the virtual environment. It is totally transparent for a user in the virtual environment due to the method; on the basis of virtualization of the dynamic library, one physical GPU is virtualized into multiple GPUs logically through the space division multiplexing and time division multiplexing technologies for a GPU pool with GPU resources as allocation objects, and the method supports that one GPU is multiplexed by multiple users.

Description

technical field [0001] The invention belongs to the field of computer architecture and virtual interface, and in particular relates to a general computing virtualization realization method based on dynamic library interception. Background technique [0002] A major challenge encountered in the development of virtual machine technology is I / O virtualization. Due to the diversity, complexity, and closure of I / O devices, it is difficult to fully and effectively use these hardware resources in the virtual machine environment, resulting in loss of virtual machine functions or performance degradation. Take the display interface card (referred to as graphics card, the main component is GPU, graphics processing unit) as an example, its main functions are two parts: graphics computing (2D, 3D) and general computing (data-intensive computing of single instruction multiple data stream type) , there are a lot of mature technologies about the virtualization of graphics computing. Howev...

AI Technical Summary

Problems solved by technology

More than 70% of Tianhe-1's computing power comes from general-purpose computing equipment. If a cloud computing framework is deployed on Tianhe-1, this part of the computing power will be difficult to utilize

[0005] However, there are still several outstanding problems in the virtualization of GPUs: first, 3D applications are highly real-time and interactive, and the peak value of inter-domain communication is extremely large, resulting in inability to respond to user commands in a timely manner or degrading image quality; The reason is that the data path is narrow. At present, GPU virtualization technology widely uses the data distribution and transmission mechanism of the Chromium project of Stanford University. This mechanism was originally designed for parallel rendering under remote conditions, and did not make any optimization for virtual machines; 3. The virtualization solution based on OPENGL can only solve the virtualization problem of graphics computing. For general computing (general purpose computing, GPU, it is relative to graphics computing)

Fourth, currently popular virtual machine technologies such as live-migration, suspend / resume, etc. do not support applications involving GPU general computing, and they will lose the ability to continue execution after migration or recovery

Images