支持国产GPU 芯片的虚拟化构建与QoS 控制研究

采用了API 转发的技术路线，通过虚拟机、虚拟机管理系统与宿主机中各模块的协作，为虚拟机提供与物理GPU 具有相同功能的显卡，在应用程序库层面实现了国产GPU 芯片的虚拟化平台。此外通过设计基于QoS 的GPU 资源自适应调度算法，为多虚拟机的资源竞争提供了公平调度保证。最后通过测试验证了基于国产GPU 芯片的虚拟化构建的可用性和基于QoS 的自适应公平性调度算法的有效性。

Research on Virtualization Construction and QoS Control Based on Domestic GPU Chips

A GPU virtualization platform is implemented for domestic GPU chips at application library level using APIforwarding, by the cooperation of the virtual machine, virtual machine management system and host. The platform providesvirtual machines with cards that have the same function of physic GPU. Besides, by designing the adaptive schedulingalgorithm based on QoS for GPU resources, the fairness between multiple virtual machines can be guaranteed. Finally,experiments have been carried out to verify the correctness of the construction of the platform based on domesticGPU chips and the fairness of the adaptive scheduling algorithm based on QoS.