计算机应用 ›› 2013, Vol. 33 ›› Issue (09): 2404-2409.DOI: 10.11772/j.issn.1001-9081.2013.09.2423

• 先进计算 • 上一篇    下一篇

面向低功耗的多核处理器Cache设计方法

方娟,郭媚,杜文娟,雷鼎   

  1. 北京工业大学 计算机学院,北京 100124
  • 收稿日期:2013-04-07 修回日期:2013-05-09 出版日期:2013-09-01 发布日期:2013-10-18
  • 通讯作者: 方娟
  • 作者简介:方娟(1973-),女,辽宁丹东人,副教授,CCF会员,主要研究方向:计算机系统结构、多核计算;
    郭媚(1988-),女,江西赣州人,硕士研究生,主要研究方向:多核计算;
    杜文娟(1986-),女,河北石家庄人,硕士研究生,主要研究方向:多核计算;
    雷鼎(1986-),男,湖北荆州人,硕士研究生,主要研究方向:多核计算。
  • 基金资助:

    国家自然科学基金资助项目;北京市教委科技计划项目

Low-power oriented cache design for multi-core processor

FANG Juan,GUO Mei,DU Wenjuan,LEI Ding   

  1. College of Computer Science, Beijing University of Technology, Beijing 100124, China
  • Received:2013-04-07 Revised:2013-05-09 Online:2013-10-18 Published:2013-09-01
  • Contact: FANG Juan
  • Supported by:

    Low-power Research for Shared Cache Multicore Processors;Research of Storage Technology for Multicore Processors

摘要: 针对多核处理器下的共享二级缓存(L2 Cache)提出了一种面向低功耗的Cache设计方案(LPD)。在LPD方案中,分别通过低功耗的共享Cache混合划分算法(LPHP)、可重构Cache算法(CRA)和基于Cache划分的路预测算法(WPP-L2)来达到降低Cache功耗的目的,同时保证系统的性能良好。在LPHP和CRA中,程序运行时动态地关闭Cache中空闲的Cache列,节省了对空闲列的访问功耗。在WPP-L2中,利用路预测技术在Cache访问前给出预测路信息,预测命中时则可用最短的访问延时和最少的访问功耗完成Cache访问;预测失效时,则结合Cache划分策略,降低由路预测失效导致的额外功耗开销。通过SPEC2000测试程序验证,与传统使用最近最少使用(LRU)替换策略的共享L2 Cache相比,本方案提出的三种算法虽然对程序执行时间稍有影响,但分别节省了20.5%、17%和64.6%的平均L2 Cache访问功耗,甚至还提高了系统吞吐率。实验表明,所提方法在保持系统性能的同时可以显著降低多核处理器的功耗。

关键词: 片上多核处理器, 二级缓存, 动态划分, 低功耗, 性能

Abstract: This paper proposed a Low-Power oriented cache Design (LPD) of Level 2 (L2) cache for multi-core processors. LPD considered three different ways to reduce the power consumption while promising the best performance: Low Power oriented Hybrid cache Partition algorithm (LPHP), Cache Reconfiguration Algorithm (CRA), and Way-Prediction based on L2 cache Partition algorithm (WPP-L2). LPHP and CRA closed the columns that were not in use dynamically. WPP-L2 predicted one appropriate way before cache accesses, which could save the access time, so as to save power. These three methods of LPD saved power consumption by 20.5%, 17% and 64.6% on average over the traditional Least Recently Used (LRU) strategy with improvement of the throughput and little influence on the runtime of programs. The experimental results show that this method can reduce the power of multi-core processors significantly and maintain the system performance.

Key words: Chip Multi-core Processor (CMP), Level 2 (L2) cache, dynamic partition, low power, performance

中图分类号: