Design And Implementation Of GPU Device Virtualization Based On Desktop Cloud

In recent years,desktop cloud technology has been paid more and more attention and development.The major companies follow the trend of technology to launch their own desktop cloud products.As an important function of modern desktop application,3D graphics display is worthy of enough attention.In the desktop cloud system environment,desktop virtualization technology provides a simulation of various hardware devices for virtual machines.At present,the open source virtualization technology of virtual machine infrastructure has been mature,but the virtualization of GPU device has not fully utilized the resources of hardware GPU and thus perform not well.GPU device virtualization performance has very large development space.Based on the full study of open source desktop cloud virtualization technology,this paper analyzes the existing representative solution of using GPU device resources in the virtual machine.With reference to the advantages of VIRTIO paravirtualization technology framework,this paper designs and implements a GPU virtualization solution,making the Linux operating system virtual machine can use the host of the physical GPU device resources to achieve accelerated graphics rendering performance.The major work includes:Firstly,the research for desktop virtualization technology,including QEMU-KVM virtualization technology,SPICE protocol.Secondly,in view of the implementation of GPU in Linux virtual machine,this paper focuses on the design and implementation of VIRTIO framework and the data transmission details of SPICE and QEMU.Thirdly,as for the development of cloud platform,this paper mainly describes the design and implementation of platform architecture.