Improving temporal joins using histograms

Abstract

Histograms are used in most commercial database systems to estimate query result sizes and evaluation plan costs. They can also be used to optimize join algorithms. In this paper, we consider how to use histograms to improve the join processing in temporal databases. We define histograms for temporal data and a temporal join algorithm that makes use of this histogram information. The join algorithm is a temporal partition-join with dynamic buffer allocation. Histogram information is used to determine partition boundaries that maximize overall buffer usage. We compare the performance of this join algorithm to temporal join evaluation strategies that do not use histograms, such as a partition-based algorithm based on sampling and a partition-join using the Time Index, an index structure for temporal data. The results demonstrate that the temporal partition-join is substantially improved through the incorporation of histogram information, showing significantly better performance than the sampling-based algorithm and achieving equivalent performance to the Time Index join without requiring an index.