多核处理器共享Cache的划分算法  

Research on Shared Cache Partition Algorithm Based on Multi-Core Processors

在线阅读下载全文

作  者:吕海玉 罗广 朱嘉炜 张凤登[1] LU Haiyu;LUO Guang;ZHU Jiawei;ZHANG Fengdeng(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science&Technology,Shanghai 200093,China)

机构地区:[1]上海理工大学光电信息与计算机工程学院,上海200093

出  处:《电子科技》2024年第9期27-33,共7页Electronic Science and Technology

基  金:国家自然科学基金(71840003);上海市自然科学基金(15ZR1429300)。

摘  要:针对多核处理器性能优化问题,文中深入研究多核处理器上共享Cache的管理策略,提出了基于缓存时间公平性与吞吐率的共享Cache划分算法MT-FTP(Memory Time based Fair and Throughput Partitioning)。以公平性和吞吐率两个评价性指标建立数学模型,并分析了算法的划分流程。仿真实验结果表明,MT-FTP算法在系统吞吐率方面表现较好,其平均IPC(Instructions Per Cycles)值比UCP(Use Case Point)算法高1.3%,比LRU(Least Recently Used)算法高11.6%。MT-FTP算法对应的系统平均公平性比LRU算法的系统平均公平性高17%,比UCP算法的平均公平性高16.5%。该算法实现了共享Cache划分公平性并兼顾了系统的吞吐率。In order to optimize the performance of multi-core processors,this study deeply investigates the management strategy of shared Cache on multi-core processors,and proposes a shared Cache partitioning algorithm MT-FTP(Memory Time based Fair and Throughput Partitioning)based on the fairness of cache time and throughput rate.A mathematical model based on the fairness and throughput index is established,and the partitioning flow of the algorithm is analyzed in the proposed study.The simulation results show that the MT-FTP algorithm has excellent performance in system throughput,and its average IPC(Instructions Per Cycles)value is 1.3%higher than that of UCP(Use Case Point)algorithm and 11.6%higher than that of LRU(Least Recently Used)algorithm.The average fairness of MT-FTP algorithm is 17%higher than that of LRU algorithm,and 16.5%higher than that of UCP algorithm.This algorithm realizes the fairness of shared Cache partition and takes into account the throughput of the system.

关 键 词:片上多核处理器 内存墙 划分 公平性 吞吐率 共享CACHE 缓存时间 集成计算机 

分 类 号:TP316.2[自动化与计算机技术—计算机软件与理论]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象