使用GPU加速无线传感器网络信道仿真

针对现有无线传感器网络信道仿真难以获得高效的执行性能问题,提出并实现了一个基于Nvidia的图形处理器(GPU)+CUDA计算体系的并行信道仿真系统;研发了可驻留于GPU高速片上存储的CUDA树群,以其组织节点,并加速探测可能的发包节点;建立了完全不同于传统信道仿真的CUDA并行信道仿真引擎. 实验结果表明,该系统以高达528.73倍的加速比远胜于相应的中央处理器实现,并线性扩展于网内节点数目.

Acceleration of Simulation of Radio Channel in Wireless Sensor Networks Using GPU

Due to the computationally-intensive characteristic, the available radio channel simulators of wireless sensor networks (WSN) cannot achieve efficient execution performances. A parallel radio channel simulator using Nvidias graphics processing unit + compute unified device architecture (GPU+CUDA) parallel computing architecture is proposed. Firstly, CUDA-trees is developed, which can reside on the fast on-chip memory of GPU, so as to organize sensor nodes and facilitate detection of possible transmitters in simulation. Secondly, a CUDA parallel radio channel simulating engine is established, which is totally different from that in traditional WSN simulators. Experiments show that this simulating system greatly outperforms a central processing unit implementation with an up to 528.73 times speedup and has a linear scalability.