GPU Coprocessing for Wireless Network Simulation

Site-specific modeling of wireless communications channels has historically been too computationally intensive to incorporate into commodity network simulators. Simulation cannot accurately predict the behavior of wireless networks in real-world environments without modeling the physical channel realistically. Realistic models typically involve large amounts of floating point computation, to which modern GPUs are well suited. In this paper we demonstrate parallel radio propagation prediction in a single machine using multiple GPUs and CPU cores. We explore the tradeoffs between model accuracy and performance, and use techniques from graphical raytracing to improve the speed with which radio path loss can be computed. GPUs; raytracing; geometric optics; propagation modeling; network simulation; parallelism; acceleration; wireless; radio; path loss; threads; CUDA; kd-trees; rays; performance