GPU虚拟化环境下云游戏性能分析

Performance Analysis of Cloud Gaming in GPU Virtualization Environment

云游戏在提供高品质游戏体验的同时带来了高额的硬件成本.以往的云游戏系统为了最大化用户的游戏体验,都是直接部署在物理服务器上,导致资源不能动态分配.针对该问题,本文使用最新的GPU虚拟化技术vGPU和开源的云游戏平台GamingAnywhere搭建原型系统.在此基础上有针对性地设计实验,研究现有的GPU虚拟化技术能否支持多个云游戏共享一台物理服务器,同时分析GPU虚拟化对云游戏的延迟和画质的影响.实验结果表明:首先,vGPU技术对GPU资源有较高的利用效率,能够支持GPU密集型的云游戏;第二,vGPU技术支持多个虚拟机之间GPU资源的公平分配;第三,使用vGPU技术将软件编码卸载到GPU执行能提高编码效率,同时降低网络负载.

Cloud gaming provides high-quality gaming experience,however,it brings the high hardware cost.The existing cloud gaming systems are deployed directly on physical servers to maximize users′ gaming experience,resulting in being unable to allocate resources dynamically.To address the issue,this paper builds a prototype system with the latest GPU virtualization technology named vGPU and the open source cloud gaming platform named GamingAnywhere.On this basis,we design a series of experiments to study whether the existing GPU virtualization technologies enable multiple cloud gaming to share a single physical server,and meanwhile,to analyze the impact of GPU virtualization on the delay and image quality of cloud gaming.The experimental results show that,firstly,vGPU has a high utilization efficiency of GPU resources,and can support the GPU intensive cloud gaming;Secondly,vGPU supports fair allocation of GPU resources among multiple virtual machines;Thirdly,by offloading software video codec to GPU using vGPU,the encoding efficiency can be improved,and network load be reduced.