虚拟化环境下GPU调度算法的研究

Research on GPU Scheduling Algorithm in Virtual Environment

在光线跟踪算法中,传统上通过像素递归深度对场景进行划分,然后必须每一个任务分配一个GPU服务器节点服务器,但是这种静态任务分配的算法在虚拟化环境中没办法实现对GPU资源的有效利用,而且一旦有服务器节点宕机就会导致场景绘制失败。对此,提出一种基于hash思想的解决方案,有效的解决该问题。

The scenes are divided by pixel recursion depth in ray tracing algorithms traditionally,in this way,one GPU node server must be assigned to each task.But this static GPU allocation algorithm can not effectively utilize GPU resources in virtualized environment,at the same time,once a node is down,it will result in scenario rendering failure.So this paper presents a solution base on hash algorithm solving the problem effectively.