Parallel radiosity on a cluster of workstations

A parallel version of progressive refinement radiosity with radiosity overshooting and adaptive surface subdivision is presented. A cluster of workstations under the operating system UNIX was chosen as parallel hardware. The whole system was organized according to the master-worker paradigm. We investigated how to adjust algorithms for run-time adaptive surface subdivision and overshooting for such parallel hardware, especially if we use a relatively slow computer network or, moreover, a network with highly varying throughput. Variation in computer performance (including the possibility that some workstations might crash) was also considered