大尺度流域水文并行计算的方法改进

Method Improvement of the Parallel Computing for the Large-scale Basin Hydrology

清华大学数字流域模型(TUD-Basin)可进行流域水沙运动过程的并行模拟,其并行计算方法使计算效率有效提高.但在实际中发现,该方法仍存在通信效率偏低和可操作性差的缺陷.本文对其进行了优化改进,一方面通过引入"中转进程",大幅减少了进程间通信消耗,显著提升了计算效率;另一方面基于匿名管道技术开发了并行计算可视化控制系统,可对计算过程中所有进程的运行状况进行实时监控,有效拓展了并行计算方法的通用性和实用性,并对岔巴沟流域自1967年5月1日0时至1967年6月1日0时的产汇流过程进行了并行计算,结果表明方法改进后的计算效率平均提高56%.

Digital basin model(TUD-Basin) developed by Tsinghua University is able to realize the parallel simulation of hydrological processes for the large-scale basin.The parallel computing method can improve the calculation efficiency obviously.However,there still exists the shortcomings of low communication efficiency and poor operability for the method.According to the situation,'Transfer Process' was designed and added to the original parallel architecture basing on MPI(Message Passing Interface).It can reduce the communication time consuming significantly.Besides,basing on the anonymous pipe technique,a visual control system was also developed that can monitor the anytime operating condition of all processes.The system expands the generality and practicability of parallel computing very efficiently.In case of Chabagou basin,the runoff process from 5/1/1967 0:00:00 to 6/1/1967 0:00:00 was simulated.The result showed that the computing efficiency of the method improved increased by an average of 56% in the case.