数据流中一种适应性查询处理机制

An Adaptive Query Processing Mechanism in Data Stream System

针对数据流中连续查询特征,本文提出一种适应性的查询处理机制,它不但能在有限时间内最大可能地输出结果元组,也可对有限的元组以最快时限输出。而此查询处理机制主要依托于基于输出速率的代价模型,此模型将不断变化的流速、谓词选择率、操作符处理时间作为代价函数变量,将输出速率作为代价模型的函数值。因此此代价模型可适应环境以及数据流本身不断变化的因素,并可作为查询计划动态选择的标准。实验证明此适应性查询处理机制最终能有效地提高输出速率、增加查询吞吐量、减少时间延迟,降低查询间内存占有量。

Aimed at properties of continuous queries, we present an adaptive query processing mechanism. It can not only optimize the time at which the last result tuple appears, but also optimize for the number of answers computed at any specified time after the query evaluation. This mechanism depends mainly on an output-rate-based cost model. Firstly, this model treats changed input rate, predicative selectivity and execution time of operators as its function variables. Secondly, it will compute output rate as function output. So this cost model can continuously adapt to variable factors of environment and data stream itself. Meanwhile, it can be a standard to select query plan. Experiment has proved that this mechanism validly increases throughput and output rate, meanwhile reduces output time latency and memory requirements.