多核编程模型运行时环境的自适应性研究

On Adaptability of the Runtime Environment for Emerging Multi-Core Programming Models

针对多核编程模型运行时环境易造成处理器核资源竞争加剧以及可扩展性较差等弊端,基于动态反馈控制思想,将资源分配、运行时控制、任务执行视为有机整体,提出了自适应协同调度模型ACSM.ACSM采用集中式与分布式相结合的协同机制,动态调节处理器核资源在不同应用负载间及其内部的分配与管理.ACSM的优势在于充分体现了多核编程模型良好的可编程性和可移植性,消除了传统多核运行时环境显式指定核数的弊端,增强了处理器核资源分配的高效性和自适应性.实验结果表明,ACSM在提高多核编程模型易用性的同时,减少了系统处理器核资源的不良竞争,提升了系统的整体性能和资源利用率.与仅依赖多核编程模型运行时环境的调度算法相比,ACSM使应用程序的运行时间缩短了近50%,并且随着应用程序数量的增加效果更加显著.

The adaptability and collaboration of the multi-core runtime system is studied to address the problems that the current multi-core runtime can easily lead to intensified competition for processor resources and the system scalability is inferior. An adaptive and collaborative scheduling model, named ACSM, is presented based upon the dynamic feedback-control principle by taking resource allocation, runtime control, and task execution as a holistic system. The ACSM dynamically reallocates and manages processor resources among and within workloads in both centralized and distributed manners. The superiorities of ACSM over the current multi-core runtime system are as follows. The ACSM maintains good programmability and portability, enhances efficiency and adaptability in processor resources allocation, and eliminates the need of explicitly specifying the number of cores. The experiment results show that ACSM greatly reduces the competition of processor resources and improves both the overall system performance and the usability of the current multi-core programming models. Comparisons with the scheduling algorithm that relies only on the original multi-core runtime show that applications of ACSM reduce the run time by about 50% or even more, especially when the system load increases.