Aggregate Sharing in Stream Databases

Abstract

We consider the problem of handling aggregate computations in a scalable fashion in stream databases. The queries of interest are sliding-window aggregate queries over a single data stream. In the na�ve approach for this problem, each query is individually processed one at a time, such that no sharing occurs among aggregates in terms of storage and answer computations. We present two new algorithms, TimeGram and OrderedSharing, both of which are designed with sharing among aggregates in mind. Our performance results show that both drastically improve the current performance of STREAM, a stream database system for which we have implemented and evaluated our two algorithms.