异构环境下基于分组的图像合成算法

Group-based Image Compositing in Heterogeneous Environments

图像合成由于需要密集的节点间通信,因此很容易成为整个并行可视化系统的瓶颈部分。传统的图像合成算法均假定环境是同构的,这种假定在异构环境下可能会造成严重的负载不均衡。针对这个问题,提出了一个异构环境下基于分组的图像合成算法。它能够根据环境的计算与网络特性更合理地进行任务分配。在ns-3网络模拟器上进行了详尽的实验评估。实验结果表明当环境存在比较大的异构性时,该方法相对于传统图像合成算法能够取得显著的加速比。

Image compositing requires dense inter-node communications and could easily become the bottleneck of a parallel visualization system. Conventional image compositing algorithms assume the environment is homogeneous and take no heterogeneity into consideration. In a heterogeneous environment, this assumption is likely to cause unreasonable load distribution among the nodes. To address this problem, we present a group-based image compositing algorithm in heterogeneous environments. It is able to distribute the workloads among the nodes according to computing and networking characteristics of the heterogeneous environments. We conduct a comprehensive evaluation on the ns-3 network simulator. Experiment results shows that when the environment has great heterogeneity, our algorithm gains significant speedup over conventional methods.