多核系统的多应用任务映射方法研究  

Research on multi-application task mapping method for multi-core system

作　　者：张伯泉[1] 费亭 宋宗峰 

机构地区：[1]广东工业大学计算机学院,广州510006 [2]广州工商学院电子信息工程系,广州528138

出　　处：《计算机应用研究》2017年第2期440-444,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61300107);广东省青年创新人才类项目(自然科学类)(2014KQNCX237);广东省教育部产学研合作专项资金资助项目(15zk0360)

摘　　要：在多核处理器系统中,多个计算任务映射到多核处理器内核的方式对于系统吞吐率至关重要。针对此问题提出一种新的多应用任务到多核的映射算法,该算法在应用到来之前预测应用的相关性能,并采用分支限界法提前为未来应用预留合适的内核几何位置。当应用真正到来时,根据预留的区域完成映射。实验结果表明,该算法相比其他传统算法,在多任务通信量的减少和多核系统的吞吐率等方面都收到了良好效果。In the muhi-core system, how to arrange the multiple computing tasks to the computing unit of multi-core system, which called mapping, is a key point to the throughput of whole system. This paper proposed a new mapping algorithm to solve this problem. This new algorithm predicted the related performance of future applications before they were coming, and select- ed the branch and bound algorithm to search and reserve appropriate geometric region for the future computing tasks. When the computing tasks comes, the whole system just needed to map them on many-core systems according to the reserved area. Experimental resuhs show that, compared with other traditional algorithms, the proposed algorithm has a good effect on reducing communications among the related computing tasks and improving the throughput of many-core computing system.

关 键 词：众核系统 多应用任务分配 吞吐率 分支限界法 

分 类 号：TP332[自动化与计算机技术—计算机系统结构]