Adaptive Scheduling Framework for Multi - Core Systems Based on the Task - Parallel Programming Model

H. M. LU, School of Computer Science and Engineering, Changchun University of Technology, Changchun 1 30012 , China, Y. J. CAO, J. J. SONG, T. Y. DI, H. Y. SUN, X. M. HAN, School of Software , Zhengzhou University, Zhengzhou 450 000, China, School of Computer Science and Engineering, Changchun University of Technology, Changchun 1 30012 , China, School of Computer Science and Engineering, Changchun University of Technology, Changchun 1 30012 , China, School of Computer Engineering, Nanyang Technological Univers ity, Singapore 639798 , Singapore, School of Computer Science and Engineering, Changchun University of Technology, Changchun 1 30012 , China
2016 Journal of Engineering Science and Technology Review  
With the rapid development of multi-core processor systems, software parallelization has become the main approach in improving the efficiency of multi-core processors. However, the most updated multi-core parallel programming models have defects, such as poor scalability and intensive competition in processor core resources. To prevent congestion of system processor core resources and to improve equal distribution of processing resource and service efficiency, the adaptive co-scheduling problem
more » ... of multi-core runtime systems was studied in this paper. First, on the basis of the online competition analysis method, a quantitative analysis of task schedulability was conducted. Second, a random work stealing strategy was combined with work stealing frequency to dynamically redistribute multi-core resources. Third, on the basis of closed-loop feedback control theory, an adaptive co-scheduling method that could obtain a dynamic perception of the degree of task parallelism was proposed, and multi-core adaptive co-scheduling system A-SYS (Adaptive SYStem) based on fine-grained task programming model was designed and implemented. Finally, the proposed framework was used to conduct performance analysis of multiple parallel tasks, and the performances of different algorithms were compared through a prototype system experiment. Experimental results indicated that the proposed adaptive scheduling method and a dynamic perception of core resources could effectively improve mutual competition between inter-core tasks and shared resources. Lower damage cost during task scheduling process, and significantly elevated the service efficiency of multi-core processors and equal distribution in resource allocation. Compared with traditional scheduling algorithm EQUI (EQUI-partitioning), A-SYS shortened the running time of application programs by nearly 50%, and as the number of application programs increased, the effect of A-SYS became more prominent. This finding is of significant reference value to performance problems caused by a continuous increase in the inner core scale of multi-core processors in the future.
doi:10.25103/jestr.096.12 fatcat:h5vgipeojfdnnbxgkw6a6fdjba