Mode Aware Stream Query Processing

Abstract

Many scientific applications including environmental monitoring, outpatient health care research, and wild life tracking require real-time stream processing. While state-of-the-art techniques for processing window-constrained stream queries tend to employ the delta result strategy (to react to each and every change of the stream sensor measurements), some scientific applications only require to produce results periodically - making the complete result strategy a better choice. In this work, we analyze the trade-offs between the delta and the complete result query evaluation strategies. We then design a solution for hopping window query processing based on the above analysis. In particular, we propose query operators equipped with the ability to accept either delta or complete results as input and to produce either as output. Unlike prior works, these flexible operators can then be integrated within one mode aware query plan - taking advantage of both processing methodologies. Third, we design a mode assignment algorithm to optimally assign the input and output modes for each operator in the mode aware query plan. Lastly, mode assignment is integrated with a cost-based plan optimizer. The proposed techniques have been implemented within the WPI stream query engine, called CAPE. Our experimental results demonstrate that our solution routinely outperforms the state-of-the-art single-mode solutions for various arrival rate and query plan shapes.