基于近似等深柱状图的数据流并行聚集算法

针对数据流并行聚集问题,提出了一种不同于关系数据和时间序列数据处理的并行聚集方法.为解决已经划分出的数据流元组无法再现的特点,提出能够感知数据流变化的采样算法对数据流采样.利用近似等深柱状图技术描述采样数据的分布特征,平均分配数据流量.使用时间聚集森林结构计算时间窗聚集.通过验证采样个数对并行聚集的影响,数据分布对近似划分向量算法性能的影响,测试数据流量与并行聚集加速比的关系,证明本算法能够高效地计算数据流聚集查询.

Parallel aggregation algorithm over data streams basedon approximate equal depth histogram

In light of parallel ag gregat ing algor ithm for data st ream, a new parallel agg reg ating st rateg yw ith relational data and temporal sequence data was put forw ard. To tackle the pro blem that part it ionedtuples w ere unable to recur over data st reams, a chang e-aw are sampling algorithm w as used to sample datast reams f irst , and an appr oximate equal depth histog ram algorithm w as used to describe data dist ributionin quer y w indow to part it io n data st reams av erag ely , and then a nov el m-order Temporal Ag greg ate forestdata str ucture w as applied to comput ing aggr eg ate in t ime w indow . It ver if ied the effect of numbers ofsampling on parallel ag gregat ion, the ef fect of data dist ribut io n on the perfo rmance o f approx imate part ition vector algorithm, and the relat io n between quant ity of data st reams and par al lel ag gr eg at ion speed ratio. Ex periment s pr ove that the proposed algorithm can pr ovide accurate answ er to parallel aggr eg ationover data str eams ef ficiently .