性能非对称多核处理器上的自适应调度

Adaptive Scheduling on Performance Asymmetric Multicore Processors

现有的性能非对称多核调度算法要么不能充分利用其体系结构而吞吐量低,要么能充分利用其体系结构但扩展性差.有些算法即使考虑了扩展性,但也局限于CPU核数目,没有考虑到任务数方面的扩展性.为了解决这些问题,作者提出了一个自适应调度算法(称为AS4AMS).在任务的每一次调度中,AS4AMS首先通过分析任务运行时的平均停驻时间得出任务的计算需求,然后根据这些需求以及各CPU核的负载情况将任务分配到合适的CPU核上运行.另外,该算法任务结束前,会不断重复上述过程以适应任务需求的不断变化.实验结果表明:与现有方法相比,所提出的方法扩展性更好并且吞吐量也更大.

Existing scheduling algorithms for performance asymmetric multicore systems either have low throughput or have bad scalability.Though scalability is considered in some algorithms,it is only confined to the number of cores,ignoring the scalability with respect to the number of tasks.To address these problems,an adaptive scheduling algorithm for performance asymmetric multicore systems,called AS4AMS,is proposed.By analyzing tasks＇ average stall time,AS4AMS obtains tasks＇ computing requirements,and then tasks are assigned to appropriate cores according to both the requirements of the tasks and the load of the cores.In addition,the above procedure is repeated to accommodate phase changes of tasks.Our experiment results show that as compared to existing algorithms,the newly proposed method delivers both higher scalability and greater throughput.