异构多核平台上基于任务划分和调度的性能评估方法

Performance evaluation based on task division and scheduling in heterogeneous multi-core systems

提出一种异构多核平台上的性能评估方法.该方法与多核任务的动态划分和调度算法相结合,可以获得给定任务集在不同硬件平台上的任务加速比.与此同时,针对不同层次的实时性需求,该算法可以给出硬件平台配置及任务划分方案.仿真实验和FPGA原型系统的测试数据表明,在任务之间存在依赖关系的情况下,该方法能够获得90%的精确度,在无依赖关系时,算法的精度为98.9%.

We present an evaluation algorithm based on task division and scheduling at run-time on heterogeneous multi-core systems. The algorithm can achieve speedups of specific tasks running on different configurations for multi-core platform within the reconfigurable hardware architectures. We build a simulation environment and implement a prototype on FPGA. Experimental results show that the algorithm can achieve 90% of theoretical speedup with dependency of different tasks and 98.9% without data dependency.