计算机工程

• 体系结构与软件技术 • 上一篇    下一篇

高通量众核并行模拟加速技术研究

方国庆1,2,李文明2,3,余洋1,张洋2,3,叶笑春2,安虹1   

  1. (1.中国科学技术大学 计算机科学与技术学院,合肥 230027;2.中国科学院计算技术研究所 计算机体系结构国家重点实验室,北京 100190; 3.中国科学院大学,北京 100049)
  • 收稿日期:2016-03-07 出版日期:2017-04-15 发布日期:2017-04-14
  • 作者简介:方国庆(1991—),男,硕士,主研方向为软件模拟技术、并行模拟框架优化;李文明,博士;余洋,硕士;张洋,博士;叶笑春,副研究员;安虹,教授。
  • 基金项目:
    国家“863”计划项目“E级超级计算机新型体系结构及关键技术路线研究”(2015AA01A301);“核高基”重大专项(2013ZX0102-8001-001-001)。

Research on Acceleration Technology in High Throughput Many-core Parallel Simulation

FANG Guoqing  1,2,LI Wenming  2,3,YU Yang  1,ZHANG Yang  2,3,YE Xiaochun  2,AN Hong  1   

  1. (1.School of Computer Science and Technology,University of Science and Technology of China,Hefei 230027,China; 2.State Key Laboratory of Computer Architecture,Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China; 3.University of Chinese Academy of Sciences,Beijing 100049,China)
  • Received:2016-03-07 Online:2017-04-15 Published:2017-04-14

摘要: 高通量应用的迅猛发展使得模拟速度成为大规模众核体系结构研究的瓶颈。为此,基于高通量众核结构模拟平台,提出一系列模拟加速技术。采用查找表方法加速指令译码,从事件调度算法、时间推进算法以及队列无锁化等角度优化并行离散事件模拟框架,以内存池管理方案提高内存管理效率。实验结果表明,与优化前方案相比,查找表、并行离散事件模拟和内存池3种加速方案在模拟速度上表现较优。

关键词: 高通量处理器, 众核模拟器, 查找表, 离散事件, 内存池

Abstract: The rapid development of high-throughput applications makes simulation speed increasingly become the bottleneck of large-scale many-core architecture research.In order to solve this problem,based on the simulation platform of high-throughput many-core architecture,a series of simulation acceleration techniques are proposed.The lookup table method is used to accelerate the decoding of instructions.In the aspects of event scheduling algorithm,time stepping algorithm and lock-free queue,the parallel discrete event simulation framework is optimized.Memory pool policy is adopted to improve the efficiency of memory management.Experimental results show that lookup table method,parallel discrete event simulation and memory pool policy achieve improvement at their corresponding stages in respect of simulation speed compared with the non-optimized ones.

Key words: high-throughput processor, many-core simulator, look-up table, discrete event, memory pool

中图分类号: