| The aggregation query is an important but costly operation in database management systems. While the aggregation in relational databases has been well studied, recently there has been a growing interest in improving the performance of computing aggregates over complex objects. Each of such objects may have a time interval, a spatial location/region, or both, as appears in temporal, spatial, and spatio-temporal databases. An aggregation query over these objects typically involves some selection condition on their time and/or spatial attributes, e.g. to aggregate over temporal records whose time intervals intersect a given time interval. A straightforward approach is to, with the aid of some index structures, locate the objects that satisfy the selection condition and aggregate their values on the fly. Such indices are general in the sense that they can be utilized not only to compute aggregates, but also to perform selection queries. However, the aggregation query performance is proportional to the number of objects satisfying the selection condition. In the worst case, to compute an aggregate, all objects in a database needs to be examined. Nevertheless, in many applications (on-line analysis, etc.), we need to compute the aggregates very fast, and to scan through all the existing objects might be too time-consuming. In this thesis, we focus on devising specialized indices for aggregation over complex objects. Our research shows that our newly designed structures have much better query performance than the existing general-index-based solutions, sometimes over a hundred times faster. In this thesis, we report our findings. |
