论文部分内容阅读
国家气象局天气组网雷达定量估测降水系统不仅拥有较大的计算量,而且具有较大的数据吞吐量,同时对实时性要求较高。如果缩短其执行时间,无疑将会带来巨大的收益。鉴于这些特点,使用VTune Amplifer XE对串行程序进行了热点分析和并行性分析,得出程序中有较多线程级并行性,从而制定了相应的并行化方案;然后使用Win32多线程和OpenMP两种技术对该程序在Intel四核处理器平台上进行了并行化。程序主要由单站处理和组网处理两部分组成。由于计算资源的限制,并行后的单站处理程序只有大约10%的性能提升,而组网处理程序则可以达到近似线性的性能提升。通过调整计算负载,并行化版本的加速比可以达到5.5。最后,可以得出该并行化方法适用于计算密集且数据吞吐量较大的一类应用。
National Weather Service weather network radar quantitative estimation precipitation system not only has a large amount of computing, but also has a large data throughput, while demanding real-time. If you shorten the execution time, no doubt will bring huge benefits. In view of these characteristics, the VTune Amplifer XE has been used to analyze the hot program and parallelism of the serial program. The result shows that there is more thread-level parallelism in the program, and the corresponding parallelization scheme is developed. Then, using Win32 multi-threading and OpenMP two The technique is parallelized on the Intel quad-core processor platform. The program is mainly composed of single station processing and network processing. Due to computational resource constraints, parallel single-station handlers have only about a 10% performance gain, while networking handlers can achieve approximately linear performance gains. By adjusting the computational load, the speedup of the parallelized version can reach 5.5. Finally, it can be concluded that this parallelization method is suitable for a class of applications that are computationally intensive and have a high data throughput.