摘要
针对Linux任务调度算法在多核系统中交互性能差的问题,提出一种分组任务调度算法GFS。根据多核系统硬件特性,自动配置物理距离近的一组CPU共享一个任务运行队列,通过平衡组内CPU对任务运行队列的访问竞争与任务迁移的代价,实现组间任务运行队列的负载均衡,减少调度延迟。通过优先调度唤醒任务,加快多核系统中交互任务的响应速度。测试结果表明,在不同任务负载下,GFS能够明显降低交互任务的平均响应时间,从而有效提高多核系统交互应用的调度性能。
To improve interactive performance of Linux in multi-core systems,this paper designs and implements an improved task scheduling algorithm named Group Fair Schedule(GFS).According to the hardware characteristics of multi-core system,GFS allows to configure a group of CPUs with close affinity to share one task run queue automatically,so that the cost of competitive access,task migration inside a group and run queue load balance between groups can be weighed,and reduces scheduling delay.GFS gives priority to awakening tasks so that interactive performance of multi-core systems is improved.Test results show that GFS decreases the average response time of interactive tasks under different background loads,and improves interactive performance of multi-core systems effectively.
出处
《计算机工程》
CAS
CSCD
北大核心
2015年第2期36-40,46,共6页
Computer Engineering
基金
"核高基"重大专项(2009ZX01028-002-003-005)
高等学校学科创新引智计划基金资助项目(B07033)
关键词
多核系统
调度算法
交互性能
自动配置
唤醒任务
负载均衡
multi-core system
scheduling algorithm
interactive performance
automotive configuration
awakening task
load balance