Paper Details

Prefetch Data Management for Parallel Particle Tracing

Authors: Tidmus, Jonathan, Miles, Roger, Chalmers, Alan G.

Abstract:

The particle tracing method uses a stochastic approach for global illumination computation of three-dimensional environments. As with many graphics techniques the computation associated with the image generation is complex. Parallel processing offers the potential of solving the computational complex particle tracing more rapidly. Distributed memory parallel systems are scalable and readily available. However, large environmental models are often bigger than individual node storage capabilities requiring data management to distribute and control the movement of environmental data as computation proceeds. Prefetch data management attempts to reduce idle time associated with remote data fetches by anticipating the latency and requesting required data items prior to their actual use during computation. This paper demonstrates how attention to work division and supply coupled with prefetch data management can be utilised to minimise overheads associated with a parallel implementation and reduce overall image synthesis time.

Proceedings:

Proceedings of WoTUG-20: Parallel Programming and Java, André W. P. Bakkers, 1997, pp 130 - 137 published by IOS Press, Amsterdam

This record in other formats: