面向多请求的并行体绘制算法

Multiple request oriented parallel volume rendering algorithm

为了提高并行体绘制算法在处理连续多个绘制请求时计算资源的利用率,提出了一种面向多请求的并行体绘制算法。该算法从两个层次并行处理多绘制请求,首先是不同请求之间的并行计算,其次是对单个请求的并行绘制。根据请求和节点的数目,提出了一种合理的任务分配策略,不仅及时响应用户请求,而且降低了连续多请求的总绘制时间。为便于负载平衡的实现,采用动态资源调度策略。在清华高性能集群计算机系统上实现了该算法。用32个处理器处理16个请求时,与现有其他算法相比,总绘制时间减少了约9.676s,响应延迟只增加了0.378s。

A multiple request oriented parallel volume rendering (MOPVR) algorithm was proposed to improve the low resource usage ratio with current parallel volume rendering algorithms processing continuous multiple requests. Multiple requests were processed in parallel in two levels using the MOPVR algorithm with one level for different requests and the other level for single requests. A reasonable work distribution strategy was developed based on the number of requests and nodes, which not only responds to user requests in real time but also reduces the total rendering time of multiple requests. A dynamical resource scheduling strategy was used to balance the loads. The algorithm was implemented on the Tsinghua High Performance Cluster System. With 32 processors processing 16 requests, the total rendering time was reduced by 9.68 s compared with other algorithms, while the response time was only increased by 0.38 s.