Multi-step Parallel PNN Algorithm for Distributed-Memory Systems

Parallel system with distributed memory is a promising platform to achieve a high performance computing with less construction cost. Applications with less communications, such as a kind of parameter sweep applications (PSA), can be efficiently carried out on such a parallel system, but some applications are not suitable for the parallel system due to a large communication cost. We focus on PNN (Pairwise Nearest Neighbor) codebook generation algorithm for VQ (Vector Quantization) compression algorithm and propose a parallel version of the PNN algorithm suitable for the parallel system with distributed memory, called "multi-step parallel PNN". The multi-step parallel PNN is a modified version of the PNN algorithm that creates a different codebook than the original PNN does, thus the quality of a codebook created by using the multistep parallel PNN may be worse than that of a codebook by the original PNN. However, our experimental results show that the quality of the codebook is almost same as that of the original one. We also confirm the effectiveness of the multi-step parallel PNN by the evaluation of the computational complexity of the algorithm and the preliminary experiment executed on a PC cluster system.