一种基于频率的多核共享Cache替换算法

发布时间：2018-05-27 06:45

本文选题：多核处理器 + 共享Cache　；参考：《电子与信息学报》2014年05期

【摘要】：LRU替换算法在单核处理器中得到了广泛应用,而多核环境大都采用多核共享最后一级Cache(LLC)的策略,随着LLC容量和相联度的增加以及多核应用的工作集增大,LRU替换算法和理论最优替换算法之间的差距越来越大。该文提出了一种平均划分下基于频率的多核共享Cache替换算法(ALRU-F)。该算法将当前所需要的部分工作集保留在Cache内,逐出无用块,同时还提出了块粒度动态划分下基于频率的替换算法(BLRU-F)。该文提出的ALRU-F算法相比传统的LRU算法缺失率降低了26.59%,CPU每一时钟周期内所执行的指令数IPC(Instruction Per Clock)则提升了13.59%。在此基础上提出的块粒度动态划分下,基于频率的BLUR-F算法相比较传统的LRU算法性能提高更大,缺失率降低了33.72%,而IPC则提升了16.59%。提出的两种算法在性能提升的同时,并没有明显地增加能耗。
[Abstract]:LRU replacement algorithm has been widely used in single-core processors, and the multi-core environment mostly adopts the strategy of multi-core sharing the last stage of Cache LLC. With the increase of the capacity and coherence of LLC and the increase of the working set of multi-core applications, the gap between the theoretical optimal replacement algorithm and the LRU replacement algorithm is increasing. In this paper, a frequency-based multi-core shared Cache replacement algorithm is proposed. This algorithm keeps some of the required working sets in Cache and ejects useless blocks. At the same time, a frequency based replacement algorithm based on dynamic partitioning of block granularity is proposed. Compared with the traditional LRU algorithm, the missing rate of the proposed ALRU-F algorithm is reduced by 26.59% and the number of instructions executed in each clock cycle is increased by 13.59%. On the basis of the proposed block granularity dynamic partitioning, the frequency based BLUR-F algorithm has better performance than the traditional LRU algorithm, and the missing rate is reduced by 33.72%, and the IPC algorithm is improved by 16.59%. The proposed two algorithms have not significantly increased energy consumption while improving performance.
【作者单位】：北京工业大学计算机学院;
【基金】：国家自然科学基金(61202076) 北京市教委科技计划面上项目(KM201210005022)资助课题
【分类号】：TP332

【相似文献】