非对称多核处理器上的操作系统集成调度

摘要

Asymmetric multi-core processors (AMP) are more energy efficient than symmetric multi-core processors (SMP) and will be the mainstream of parallel computing architecture in the future. The existing researches on the problem of parallel task scheduling in operating systems (OS) on AMP assumed all cores have constant frequencies. They haven't analyzed the problem theoretically. These researches took neither the energy efficiency nor the universality of the scheduling into account. To solve this problem, a scheduling model based on nonlinear programming is proposed in this paper. Moreover, scheduling principles of comprehensively considering synchronization characteristics of parallel tasks, asymmetry and load of cores are analyzed and adhered. An integrated scheduling algorithm are also proposed based on the model. The algorithm integrated thread scheduling and dynamic voltage and frequency scaling (DVFS) in OS to improve energy efficiency. In addition, the algorithm achieved universality with a flexible parameter adjustment mechanism. It is the first algorithm to exploit thread scheduling and DVFS on AMP simultaneously. The evaluation on real platform demonstrates that the algorithm is universal for different conditions and it always outperforms other scheduling algorithms on asymmetric multi-core processors (by 24%~50%).%相对于对称多核处理器,非对称多核处理器具有更高的效能,将成为未来并行操作系统中的主流体系结构.对于非对称多核处理器上操作系统的并行任务调度问题,现有的研究假设所有核心频率恒定,缺乏理论分析,也没有考虑算法的效能和通用性.针对该问题,该文首先建立非线性规划模型,分析得出全面考虑并行任务同步特性、核心非对称性以及核心负载的调度原则.然后,基于调度原则提出一个集成调度算法,该算法通过集成线程调度和动态电压频率调整来提高效能,并通过参数调整机制实现了算法的通用性.提出的算法是第一个在非对称多核处理器上结合线程调度和动态电压频率调整的调度算法.实际平台上的实验表明:该算法可适用于多种环境,且效能比其他同类算法高24％～50％.