A Comparison of Indexed Temporal Joins

We examine temporal joins in the presence of indexing schemes. Utilizing an index when processing join queries is especially advantageous if the join predicates involve only a portion of the temporal relations. This is a novel problem since temporal indices have various characteristics that can affect join processing drastically. For example, temporal indices commonly introduce record copies to achieve better clustering of records with long intervals. We first identify such issues and show that naïve approaches do not work. We concentrate on common temporal join queries and examine various index-based algorithms for processing them. Two optimization techniques for indexed joins are proposed, namely, the balancing condition optimization and the virtual height optimization. While we use the Multiversion B+-tree as the temporal index, our results apply to other efficient tree-based temporal indices. We also compare against using a traditional R-tree as the join index and a Spatially Partitioned Join approach. Based on the temporal data characteristics and the type of temporal join query, we identify the best approach to process the join, a useful outcome for a temporal query optimizer.