LFF:一种面向大数据应用的众核处理器访存公平性调度机制
发布时间：2018-12-06 19:53
【摘要】：研究了众核处理器的访存公平性问题。针对众核处理器距离访存资源较近的处理单元拥有较大的访存带宽而造成的访存公平性问题,提出了一种面向大数据应用的众核处理器访存公平性调度机制:最少最远(LFF)优先访存。这种机制的原理如下:依据处理单元距离访存资源的距离以及处理单元访存的次数来调度访存顺序,以保证各个处理单元的公平性。首先,访问次数较少的节点被赋予更高的访存优先权。其次,在具有相同访问次数的节点中,距离更远的节点优先访存。再次,在相同距离的节点中,已被选中优先次数少的有优先级。实验评估表明,该调度机制能够有效解决众核处理器的访存公平性问题,其公平性调度效果优于FR-FCFS,PAR-BS、ATLAS。在1024核情况下,系统异步率由FR-FCFS的15.5%降低到1.89%。
[Abstract]:The problem of memory access fairness for multicore processors is studied. In view of the fairness of accessing memory caused by the larger access bandwidth of multi-core processor, the processing unit which is close to memory access resource has a lot of memory access bandwidth. In this paper, a scheduling mechanism for accessing memory fairness of multicore processor based on big data is proposed: minimum farthest (LFF) priority access. The principle of this mechanism is as follows: according to the distance between the processing unit and the memory access resource and the number of times the processing unit accesses the memory access order is scheduled to ensure the fairness of each processing unit. First, nodes with fewer accesses are given higher access priority. Second, in nodes with the same number of visits, the more distant nodes have priority access. Again, at the same distance, the node that has been selected with less priority has priority. The experimental results show that the scheduling mechanism can effectively solve the memory access fairness problem of multi-core processors, and its fairness scheduling effect is better than that of FR-FCFS,PAR-BS,ATLAS.. In the case of 1024 core, the asynchronous rate of the system is reduced from 15. 5% of FR-FCFS to 1. 89%.
【作者单位】： 计算机体系结构国家重点实验室(中国科学院计算技术研究所);中国科学院大学计算机与控制学院;数学工程与先进计算国家重点实验室;
【基金】：国家自然科学基金(61332009);国家自然科学基金创新研究群体科学基金(61521092) 国家重点研发计划课题(2016YFB0200501) 北京市科委项目(Z151100003615006)资助项目
【分类号】：TP311.13;TP332
本文编号：2366568
[Abstract]:The problem of memory access fairness for multicore processors is studied. In view of the fairness of accessing memory caused by the larger access bandwidth of multi-core processor, the processing unit which is close to memory access resource has a lot of memory access bandwidth. In this paper, a scheduling mechanism for accessing memory fairness of multicore processor based on big data is proposed: minimum farthest (LFF) priority access. The principle of this mechanism is as follows: according to the distance between the processing unit and the memory access resource and the number of times the processing unit accesses the memory access order is scheduled to ensure the fairness of each processing unit. First, nodes with fewer accesses are given higher access priority. Second, in nodes with the same number of visits, the more distant nodes have priority access. Again, at the same distance, the node that has been selected with less priority has priority. The experimental results show that the scheduling mechanism can effectively solve the memory access fairness problem of multi-core processors, and its fairness scheduling effect is better than that of FR-FCFS,PAR-BS,ATLAS.. In the case of 1024 core, the asynchronous rate of the system is reduced from 15. 5% of FR-FCFS to 1. 89%.
【作者单位】： 计算机体系结构国家重点实验室(中国科学院计算技术研究所);中国科学院大学计算机与控制学院;数学工程与先进计算国家重点实验室;
【基金】：国家自然科学基金(61332009);国家自然科学基金创新研究群体科学基金(61521092) 国家重点研发计划课题(2016YFB0200501) 北京市科委项目(Z151100003615006)资助项目
【分类号】：TP311.13;TP332
【相似文献】
相关期刊论文 前1条
1 王济勇,赵海,林涛,王小英,王金东,韩光洁;定时器驱动的RM调度机制建模及其性能优化[J];计算机学报;2005年02期
相关硕士学位论文 前4条
1 刘磊;异构五核XDSP调度机制的设计与实现[D];国防科学技术大学;2013年
2 仇文娟;云计算中依赖任务动态并行调度机制的研究[D];大连理工大学;2011年
3 闫英辉;云计算架构及调度机制的研究[D];大连理工大学;2011年
4 侯旭萌;基于BitTorrent的流媒体缓存及调度机制研究[D];燕山大学;2012年
，本文编号：2366568
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2366568.html
