论文部分内容阅读
具有高通量特征的大数据应用已成为目前数据中心的主流应用,这些应用在传统处理器平台上的运行效率不高,原因之一是任务调度的低效。针对高通量应用的一些典型特征以及现有任务窃取算法的不足,该文提出一种程序行为和环境感知的任务调度机制,通过软硬件结合实现了处理器核的分区管理和任务的分级调度,减小了不同应用之间因争用共享资源对性能产生的不利影响,同时利用线程相似度高的特点提高指令缓存的命中率,从而提升系统的整体吞吐率。初步的模拟评估表明:该算法在混合负载情况下性能明显优于现有算法的,在测试的混合负载中平均优于现有算法20%。
Big data applications with high-throughput features have become mainstream applications in data centers today. One of the reasons that these applications run inefficiently on traditional processor platforms is inefficient task scheduling. Aiming at some typical characteristics of high-throughput applications and the existing algorithms for task stealing, this paper proposes a task scheduling mechanism based on program behavior and environment awareness. By combining software and hardware, partition management and task scheduling are implemented , Reducing the adverse impact of competing shared resources on performance between different applications, and improving the overall throughput of the system by utilizing the high similarity of threads to increase the hit rate of the instruction cache. The preliminary simulation results show that the proposed algorithm performs significantly better than the existing algorithms under mixed load conditions and averages better than 20% of the existing algorithms in the tested mixed load.