摘要
为了解决雷达数据处理系统数据量日益增大,计算能力逐渐不足的问题,提出两种并行处理方法。第一种方法是对数据处理各步骤中的循环采用多个线程并行处理,属于细粒度并行;第二种方法是根据雷达数据的局部性特征,把雷达探测空域按照径向距离划分成多个部分,由多个子任务并行处理,属于粗粒度并行。实验结果显示,4线程细粒度并行雷达数据处理架构性能是原来的3倍,4任务粗粒度并行架构性能是原来的5倍,证明并行处理技术在雷达数据处理中的有效性,并且任务级的粗粒度并行架构更适合雷达数据处理。
In order to solve the shortage of computing ability in radar data process with increasing computation, two par- allel process methods are proposed based on the analysis of radar data process. The first method, the loop operations in every step of data process are processed by multiple threads, which is fine-grained parallel; the second method, radar detecting air- space is divided into several parts by radial range based on the locality feature of radar data, the divided parts are processd by multiple tasks, which is coarse-grained papallel. The experiment result shows that, the performance of 4 threads fine- grained parallel radar data process architecture is almost 3 times of serial data process, and the performance of 4 tasks coarse-grained parallel is nearly 5 times, which proves parallel process is effective in radar data process and tasks level coarse- grained parallel architecture is more suitable.
出处
《计算技术与自动化》
2013年第2期109-114,共6页
Computing Technology and Automation
