We present a hierarchical master-slave architecture for performing parallel raytracing algorithms that can support a large population of participating clients and at the same time maintain fault tolerance as the application level. Our design allows for scalability with minimal data redundancy and maximizes the utilization of each client involved in the raytracing process. Our results show that this three-layer system can survive any type or number of client failures, and any non-concurrent server failures, while maintaining a near linear increase in performance with the addition of each new processing client.

• James Skorupski, Ben Weber, Mei-Ling L. Liu: A High Population, Fault Tolerant Parallel Raytracer. SEDE 2006: 122-127

• Download the paper here.