Energy-Efficiency Schedule Algorithm for Real-Time Tasks with Multiple Parallel Segments on Multicore Systems

• Main Authors: Lin, Jian-Xing, 林建興
• Other Authors: Kuo, Chin-Fu
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/jbgwkk

碩士 === 國立高雄大學 === 資訊工程學系碩士班 === 106 === In a multiple-processor system, there are many applications with real-time requirements. Some applications need a lot of computing capacity from the processors. Such an application can not satisfy its real-time requirement if it is executed on only a processor. If the application is parallelized to be executed on multiple processors, the real-time requirement will be satisfied. In this thesis, we investigate the problem where there is a task set including parallel tasks and sequential tasks executing in a multiple-processor system. To schedule the tasks and energy-efficiently execute them, we propose a systematic mechanism to do parallelization for parallel tasks and assign subtasks to processors. A parallel task consists of several sequential segments and parallel segments. A sequential segment is executed only with a thread while a parallel segment can be executed with multiple threads. We try to derive a suitable parallel combination for the segments of a parallel task and divide it into several subtasks. Each segment has its corresponding subtasks. The Equal Flexibility mechanism is used to set the relative deadline for the subtasks. Then, the execution order constraints are transformed into timing constraints. Each subtask has its feasible interval and it only can be executed in the interval. If all the subtasks of a parallel task meet their deadlines, the task can meet its end-to-end deadline. Considering that the feasible interval of the subtasks of different segments of a parallel task have no overlaps, we propose a computing-capacity share mechanism to assign the subtasks of different segments into the same processor as a share group. The subtasks in a share group can share the computing capacity which is equal to the maximum density of the subtasks. If a subtask can not join a share group, the Worst Fit (WF) mechanism is adopted to assign the subtask for energy efficiency. The related properties are also proposed. Finally, we build a simulation model to verify our proposed mechanism. We compare the compared mechanism in term of schedulability, energy consumption, and the combination of schedulability and energy consumption. The experimental results demonstrate that the performance of the proposed mechanism is better than the compared mechanisms.