Fault-Tolerance of Parallel Volume Rendering on Cluster of PCs

In this paper we address a very important issue in parallel rendering systems, reliability. Distributed systems, such as clusters of PCs, are low-cost alternatives for running parallel rendering systems. However, distributed systems are usually not reliable, machines can fail during the rendering process, resulting in incomplete final images. Therefore, our goal is to take advantage of specific features of the parallel rendering applications, like tile-based computation, to include mechanisms to dynamically detect machine failure and automatically process tasks retrieval, with low overhead and no extra hardware. We developed three different parallel rendering systems, all based on the Parallel ZSweep algorithm[5], to provide fault-tolerance in different ways. Our experimental results show that the three systems present a small overhead to detect the failures, and when a failure occurs, the redistribution of the work does not degrade the system performance. We conclude that it is possible to provide fault-tolerance at low-cost in a cluster of PCs.