BDSim:面向大数据应用的组件化高可配并行模拟框架

发布时间：2018-01-10 02:20

本文关键词：BDSim:面向大数据应用的组件化高可配并行模拟框架　出处：《计算机学报》2015年10期 　论文类型：期刊论文

【摘要】：大规模并行模拟是研究大数据体系结构的重要方法,对大数据应用及众核体系结构的发展有着不可替代的推动作用.然而,目前的模拟技术不能满足大数据体系结构研究的需求,主要体现在模拟速度慢、配置过程复杂以及可扩展性差等方面.为了解决此问题,评估面向大数据应用的高通量众核体系结构的性能与功耗,该文提出了面向大数据应用的并行模拟框架——BDSim.该框架基于组件化思想,将功能组件与框架服务单元组成并行功能单元,并可根据负载情况,自由配置组件与框架服务单元之间的映射关系.为了提高组件之间的通信和同步效率,该文提出了一种非阻塞无锁通信优化方法,和一种CMB保守同步算法的优化算法——NMTRT-CMB同步算法.模拟不同并发规模的基于2D-Mesh网络的众核系统的实验结果表明,与基于锁的并行通信方法相比,框架采用的非阻塞无锁通信优化方法可以提高并行模拟速度约10%,该算法与CMB同步算法相比,NMTRT-CMB同步算法可以减少空消息数量达90%以上.
[Abstract]:Large-scale parallel simulation is an important method to study big data architecture, which can not be substituted for the application of big data and the development of multi-core architecture. The current simulation technology can not meet the needs of big data architecture research, mainly reflected in the slow simulation speed, complex configuration process and poor scalability. In order to solve this problem. To evaluate the performance and power consumption of high-throughput multi-core architecture for big data application, a parallel simulation framework for big data application, BDSim-based, is proposed in this paper. The functional components and the frame service unit are composed of parallel function units, and the mapping relationship between the components and the framework service units can be freely configured according to the load situation. In order to improve the communication and synchronization efficiency between the components. In this paper, a non-blocking and lock-free communication optimization method is proposed. And an optimization algorithm of CMB conservative synchronization algorithm, NMTRT-CMB synchronization algorithm. The experimental results show that the multi-kernel system based on 2D-Mesh network with different concurrency scale is simulated. Compared with the Lock based parallel communication method, the non-blocking and lock-free communication optimization method used in the framework can improve the parallel simulation speed by about 10%, which is compared with the CMB synchronization algorithm. NMTRT-CMB synchronization algorithm can reduce the number of empty messages to more than 90%.
【作者单位】：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;中国科学院大学;华为技术有限公司中央研究院;高效能服务器和存储技术国家重点实验室;数学工程与先进计算国家重点实验室;
【基金】：国家“九七三”重点基础研究发展规划项目基金(2011CB302501) 国家“八六三”高技术研究发展计划项目基金(2012AA010901,2015AA011204) “核高基”国家科技重大专项基金项目(2013ZX0102-8001-001-001) 国家自然科学基金(61173007,61204047,61332009)资助~~
【分类号】：TP332
【正文快照】： 李文明1),2)叶笑春1)张洋1),2)宋风龙3)王达1)唐士斌4)范东睿1)谢向辉5)1)(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)2)(中国科学院大学北京100049)3)(华为技术有限公司中央研究院北京100085)4)(高效能服务器和存储技术国家重点实验室北京100085)5)(

【参考文献】