一种并行性能模拟中LP到PP的高效映射方法  

An Efficient Mapping Method from LP to PP in Parallel Performance Simulation

作　　者：方建滨[1] 徐传福[1] 车永刚[1] 翁玉芬[1] 王正华[1] 

机构地区：[1]并行与分布处理国家重点实验室,湖南长沙410073

出　　处：《计算机工程与科学》2009年第A01期218-220,233,共4页Computer Engineering & Science

基　　金：国家863计划资助项目(2007AA01Z116);国家自然科学基金资助项目(60603055)

摘　　要：LP到PP的高效映射是加速并行性能模拟的关键技术之一。针对交互规则的并行应用程序,设计映射生成方法A2-LP3M从Trace中提取LP间的交互模式,以宿主机物理进程间通信最小化为目标,兼顾计算负载平衡,从循环块映射中选取合适的映射方式。实验表明,相对于常规映射方法,A2-LP3M减少并行模拟时间最多可达16.2%。Efficient mapping from LP to PP is one of key technologies to accelerate parallel performance simulation. Aiming at minimizing the communications between host physical processes and balancing computational load, we present a method to get communication pattern of logical processes for parallel applications that have regular communication pattern. Then we choose the rational mapping among block-cyclic mappings. Experimental results show that A^2-LP^2 M can reduce simulation time up to 16. 2% compared with the regular mapping method.

关 键 词：并行性能模拟 映射 通信优化 负载平衡 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]