Performance Analysis of NVIDIA GPU Virtualization in NARI Desktop Cloud
Zhao Wang, Miao Jingwen, Jun Yu, Zhu Guangxin · 2019
As a typical cloud computing application model, desktop cloud has become a good solution for power companies. How to improve GPU utilization have become a key issue in the use of Virtual Graphics Processing Unit (vGPU) in desktop cloud. To solve this problem, this paper builds a prototype system to analyze the feasibility of using GPU sharing in NARI desktop cloud (NCloud-D). On this basis, we design an experiment to compare performance between virtual desktops using vGPU and the physical machine. We analyze the FPS of one vGPU virtual desktop which is assigned different amount of graphics frame buffer. In addition, we sum FPS of concurrent vGPU virtual desktops. Our experimental results show that virtual desktop using vGPU in NCloud-D has the same performance as the physical machine, which can basically meet the business requirements of graphics acceleration processing. This paper can be used as basis for engineering applications of vGPU in the desktop cloud.