为了提高异构多核处理器的性能和资源利用率,研究了优化异构多核处理器的程序调度方法。针对异构多核处理器的特点,提出了一种基于神经网络的低开销程序性能预测的调度模型。该调度模型根据程序固有特征预测各个程序在不同处理器核上的性能,然后根据性能预测找出程序与处理器核之间的最优匹配方案进行调度。试验证明,该调度模型对于异构多核处理器的性能和能效都取得了很好的提升效果,超过了现有的轮转调度、抽样调度和性能影响评估(PIE)调度。相比于轮转调度,该调度模型在处理器性能和能效上分别取得了13.64%和10.78%的提升。
The optimization of the program scheduling for heterogeneous multi-core processors was studied to improve the processors' performance and resource usage, and a new scheduling model based on neural networks' low cost prediction of program performance was proposed in view of the characteristics of heterogeneous multi-core processors. The scheduling model predicts the performance of each program on different cores according to the inherent program characteristics, and then makes the best program-core matching scheme based on the program predictions for program scheduling. The experimental results demonstrate that the proposed scheduling model outperforms the existing models of the round robin scheduling, the sampling-based scheduling and the performance impact estimation (PIE) scheduling in both performance and energy efficiency. For example, compared with the round robin scheduling, the performance and the energy efficiency of the proposed model increased by 13.64% and 10.78%, respectively.