A parallel implementation of nearest neighbor analysis based on GPGPU

Nearest neighbor analysis is one of the classic methods to find out the tendency of the observed point dataset. With the explosion of spatial data, conventional implementation of nearest neighbor analysis cannot present high performance towards large amount of dataset. So in this paper, a parallel implementation of nearest neighbor analysis is proposed, with parallelization of computing the nearest neighbor distance of each point. Compared with CPU, now GPU can provide more powerful capacity of processing floating point operations and has more multiprocessors for parallel processing. So we develop the parallel program of nearest neighbor analysis with CUDA (Compute Unified Device Architecture) in terms of GPGPU (General-Purpose computing on Graphics Processing Units). In our experiments, when the number of points is large, the speedup of the parallel implementation can achieve more than 10, compared with the conventional implementation in CPU.