一种均衡可扩展计算机体系结构分布式模拟方法  

Scalable and Evenly Distributed Simulation Method for Computer Architecture

作　　者：徐传福[1] 车永刚[1] 王正华[1] 彭宇行[1] 

机构地区：[1]国防科学技术大学计算机学院,湖南长沙410073

出　　处：《软件学报》2014年第8期1844-1857,共14页Journal of Software

基　　金：国家自然科学基金(11272352);国家高技术研究发展计划(863)(2007AA01Z116);国家重点基础研究发展计划(973)(2009CB723803)

摘　　要：分布式并行模拟是提高体系结构模拟速度的有效技术手段之一.首先,建立了分布式并行模拟的通用性能分析模型,并对典型系统的并行加速比、并行效率等性质进行了理论分析,得出了一些有用的结论.在此基础上,提出了均衡可扩展分布式并行模拟方法 SEDSim(scalable and evenly distributed simulation).SEDSim针对模拟节点负载不均衡问题,提出了开销模型指导的指令区间均衡分割和分配策略CoMEPA(cost model guided evenly partition and allocation);针对分布式并行模拟与非连续、任意数量抽样模拟区间的高效集成,提出了基于最小等价距离(minimum equivalent cost,简称MinEC)的指令区间分配策略MinEC.基于sim-outorder实现了SEDSim,采用SPEC CPU2000中的部分程序对其速度和精度进行了测试,理论分析和测试结果均表明了SEDSim的优势:相对于常用的方法或策略,CoMEPA和MinEC分别能够获得多达约1.6倍和1.4倍的性能提升.Distributed simulation is an effective method to improve simulation speed for computer architecture. In this paper, a general performance model for distributed simulation is established and then some typical distributed simulation systems are analyzed based on the model. The analysis results in some important conclusions about parallel speedup and parallel efficiency for distributed simulation. Next, a scalable and evenly distributed simulation （SEDSim） approach is presented. SEDSim adopts a cost model guided even partition and allocation （CoMEPA） policy for benchmark program instructions to enhance load-balance among parallel simulation nodes. An allocation policy based on minimum equivalent cost （MinEC） is also designed to efficiently integrate arbitrary number of discrete sampling intervals in SEDSim. The study implementes SEDSim based on sim-outorder and evaluates its speed and accuracy using Benchmark programs from SPEC CPU2000. Both theoretical analysis and testing results validate some advantages of SEDSim approach. Compared with existing methods, CoMEPA and MinEC can achieve a speedup of about 1.6 and 1.4 respectively.

关 键 词：计算机体系结构 分布模拟 负载均衡 可扩展 

分 类 号：TP306[自动化与计算机技术—计算机系统结构]