摘要
提出了基于直方图的降载策略,能很好地减缓在过载发生时系统性能的下降.降载的目标在于删除过载数据的同时尽可能地保持数据流的特征.为了处理大量被延迟的数据,构建了一种塔形矩阵的数据存储结构,利用其对过载数据分桶,每桶提取一个代表数据并删除该桶中其余数据,将每个桶的代表数据组成新的数据流参与查询操作.实验结果表明:这种降载方法能有效减少系统负担,生成的新数据流参与数据流查询后所得查询结果错误率较低,其性能优于其他已有算法.
A technique for gracefully degraded performance via load shedding was examined,which was histogram strategy.The goal was to load shedding,as much as possible to preserve feature and improve the latency of the overloaded system.In order to deal with large amount of delayed data,a pyramid metric storage structure was conceived.With this data structure,how to split the data stream to buckets,extract the representative data from these buckets and drop the remaining data from every bucket could be found out.New data streams involved in the data streams query was generated by these representative data.Several experiments on data streams were performed.The results prove that the new method do well in percentage error of query result,and this new load shedding method has super performance compared with the existing strategies.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2014年第9期24-27,共4页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(61173049)
关键词
数据流
降载
直方图
塔形矩阵
过载
data streams
load shedding
histogram
pyramid metric
overload
作者简介
魏霞（1976-），女，博士，E—mail：119075307@qq．com．
