One of the most important uses of aggregate queries over data streams is sampling. Typically, aggregation is performed over sliding windows where queries return new results whenever the window contents change, a concept referred to as a continuous query. Existing data models and query languages for streams are not capable of expressing many practical user-defined samplings over streams. To this end we propose a new data stream model, referred to as the sequence model, and a query language for specifying aggregate queries over data streams. We show that the sequence model can readily express a superset of the aggregate queries expressible in the previously proposed time-based data stream model, thus providing a declarative and formal semantics to understand and reason about continuous aggregate queries. Defined on top of the sequence model, our query language supports existing sliding window operators and a novel frequency operator. By using the frequency operator one is capable of expressing useful sampling queries, such as queries with user-defined group-based sampling and nested aggregation over either the input stream or the result stream. Such capabilities are beyond those of previously proposed query languages over streams. Finally, we conduct a preliminary experimental study that shows our language is effective and efficient in practice.
展开▼
