【作者】 韩东红； 王国仁； 周锐； 公丕臻； 王剑； 杜钰；

【Author】 Han Donghong,Wang Guoren,Zhou Rui,Gong Pizhen,Wang Jian,and Du Yu (School of Information Science and Engineering,Northeastern University,Shenyang 110004)

【机构】 东北大学信息科学与工程学院；

【摘要】 数据流管理系统中的查询处理需要满足各种服务质量(QoS)要求,其中延迟是用户非常关注的QoS参数.如何在系统资源有限、数据输入速度及特征可变,且单个元组的CPU处理代价动态变化的情况下,使系统具有自适应的能力以确保高质量的QoS,是迫切需要解决的问题.为此,提出了一种全面考虑CPU处理能力及内存容量的QoS自适应框架,并运用基于控制理论的PID控制器管理内存负载,超载时进行卸载处理;同时提出有效的清洗策略和队列调度策略,保证系统在动态环境中具有良好的自适应性.实验结果表明,该框架具有稳定性好、健壮性强的优点,在错失率相同的情况下,卸载率、系统资源利用率等几方面性能均优于其他现有方法.更多还原

【Abstract】 Query results in data stream management system should meet the requirements of all kinds of QoSs,of which tuple delay is an important one that most users care about.Consequently,it is a big challenge to adjust the system to provide high QoS quality in the case of limited resources,variable stream arrival rates and fluctuating per-tuple processing cost.In this paper,a QoS adaptation system framework taking both CPU and memory into account is proposed and PID controller is utilized to shed load when memory is about to overflow.Meanwhile effective scheduling strategies and clean method are also applied in memory management to extend the adaptability of the system in dynamic environments.The experiments show that not only is the system stable and robust,but also outperforms other existing works on resource utilization and shedding rate under a given target deadline miss ratio.更多还原