A data allocation method in multi-processors task scheduling procedure

Chao Wang,Wei Liu,PeiYuan Yuan

DOI: https://doi.org/10.1007/978-3-642-25905-0_32

2011-01-01

Abstract:Traditional task scheduling algorithms concentrate on the optimization of the speedup in parallel processing, where the data allocation procedure which maps the data operation of each task to the physical memory in processors is usually omitted. In this paper, the data allocation problem in task scheduling procedure is analyzed and explored. An allocating method is proposed, which consists of three steps. Circled digit oneEstablish the memory model for all the compute tasks and IPC tasks; Circled digit twoConstruct the set SE for the determination of the executing priority relations between tasks through the DFS(Deep First Search) method; Circled digit threeAllocate data buffers of tasks with different executing priority to the same physical memory areas to achieve memory reusing. The experiment shows that by make use of the executing priority relations, different tasks could accessing the same physical memory areas through sharing-timing mode, which reduces the memory usage effectively. © 2011 Springer-Verlag.