第七章 分层调度模型.pdfVIP

众核处理器系统的分层调度模型研究 1 1,2 1 1 曹仰杰 , 钱德沛 , 伍卫国 , 董小社 (1. 西安交通大学电子与信息工程学院, 710049, 西安; 2. 北京航空航天大学计算机学, 100191, 北京) 摘 要: 针对众核处理器核间并行规模日趋增大、性能难以充分发挥等问题，提出了一种兼顾公平与效率 的分层在线调度模型(McHS) 。基于运行时环境，McHS 在不同调度层次引入具有不同特性的调度算法以实 现处理器核资源的高效管理与分配。在上层，待运行任务依据其特性动态组织成不同的任务簇，簇间采用 均衡算法 EQUI 进行资源分配；下层调度簇内，任务间资源分配策略采用基于乘性增减反馈机制的自适应 调度算法。利用决策优化理论的在线竞争分析方法，证明了 McHS 中任务执行时间代价的在线竞争比为 2 。实验分析表明，与多层 EQUI 及动态窃取算法相比，McHS 提高了核资源的利用率，在同样应用负载情 况下，任务集的运行时间减少了近15%和9% 。 关键词: 众核处理器；分层调度；在线调度；竞争分析 中图法分类号: TP301 文献标识码: A 文章编号：0372-2112 （） Research on Hierarchical Scheduling Model for Many-core Systems Abstract: Aiming at addressing the increasing difficulty of efficiently using large number of cores in many-core processors, a hierarchical scheduling model, named MsHS, is proposed with emphasis on fairness and efficiency. Based on many-core runtime environment, McHS can efficiently manage and allocate computing resources by integrating different scheduling algorithms at different levels. At the high level, parallel jobs are dynamically organized into groups according to their characteristics and the EQUI scheduling algorithm is adopted to allocate computing resource among groups. At the low level, a feedback-driven adaptive scheduling algorithm, based on multiplicative-increase multiplicative-decrease policy, is implemented to dynamically adjust resource allocation within groups. Competitive analysis commonly used in decision optimization theory shows that McHS achieves 2- competitiveness ratio with respect to the execution time of parallel jobs. The experimental results demonstrate that compared with hierarchical EQUI and work-stealing algorithms, McHS not only improves the computing resource utilization significantly but also reduces the exec