Optimizing In-Order Execution of Continuous Queries over Streamed Sensor Data

Abstract

The widespread use of sensor networks in scientific and engineering applications leads to increased demand on the efficient computation of the collected sensor data. Recent research in sensor and stream data systems adopts the notion of sliding windows to process continuous queries over infinite sensor readings. Ordered processing of input data is essential during query execution for many application scenarios. In this paper we present three approaches for ordered execution of continuous sliding window queries over sensor data. The first approach enforces ordered processing at the input side of the query execution plan. In the second approach we utilize the advantage of out-of-order execution to optimize query operators and enforce an ordered release of the output results. The third approach is adaptive and switches between the first and second approaches to achieve the best overall performance with current input arrival rates and level of multiprogramming. We study the performance of the proposed approaches both analytically and experimentally and under a variety of conditions such as the asynchronous arrival of input data, and various levels of multiprogramming. Our performance study is based on an extensive set of experiments using a realization of the proposed approaches in a prototype stream query processing system.