Toward Accelerated Line-Of-Sight Intervisibilty Calculations Using Clusters Of Gpus

The processing power of graphical processing units in recent years has been increasing at a rate that exceeds the so-called "Moore's Law" for general purpose CPUs, while the prices for these GPUs has dropped precipitously. Beginning in the late 1990's, researchers realized that this enormous processing power could be used to solve problems other than simply image generation. Almost in parallel to these developments, other researchers began using dedicated networks of commodity computers and supporting network hardware to construct low-cost supercomputers (Beowulf clusters) capable of solving particular problems formerly requiring much more expensive proprietary supercomputers. In this paper we examine combining these two concepts with the eventual intention of rapidly accelerating intervisibility calculations for CGF and constructive simulations. We present initial experimental results on the computation time and scalability of using clustered GPUs to calculate intervisibility over densely populated terrain databases. We also discuss intervisibility correlation between CGF and GPU-based approaches, and present an example of differences in intervisibility calculations that are inherent in the different systems.