基于实例学习的并行负荷分配中的训练实例选择问题

Training Example Selection in Instance-Based Learning Parallelization

基于实例学习的可适应性并行任务负荷分配算法能根据应用程序的静态特征估计其运算负荷,选定好的任务负荷分配方案使其多线程并行接近甚至达到最优,它具有低成本和高效率的特点.通过一系列实验,分析研究训练实例的选择对基于实例学习优化的效果的影响,从中总结一些有益的经验,以便进一步提高算法性能.

A learning-based approach uses static program features to estimate Java program's workload before allocate parallel workload among Java threads in order to achieve optimal higher performance. This paper analyses the impact training example selection can make on the efficiency of this approach, based on the results of a series of experiments.