Index based processing of semi-restrictive temporal joins

Temporal joins are important but very costly operations. While a temporal join can involve the whole time (and/or key) domain, we consider the more general case where the join is defined by some time-key rectangle from the whole space (i.e., when the user is interested in joining portions of the-usually large-temporal data). In the most restrictive join, objects (within this rectangle) are joined together based on key equality and interval intersection. This paper concentrates on semi-restrictive joins, i.e., when either the key equality (equi-join) or the interval intersection (time-join) predicates are used. Given the large relations created by the ever increasing time dimension, we assume that each temporal relation is indexed and examine efficient ways to process semi-restrictive temporal joins. Utilizing an index is helpful since it directs the join towards the objects that are within the time-key rectangle. A straightforward approach is to perform an unsynchronized join. An index selection query on each relation identifies all objects within the time-key rectangle which are then joined. Although simple, this approach ignores the data distribution in the other relation. Instead, in a synchronized join, both indices are concurrently traversed as the join is computed. Synchronized semi-restrictive join algorithms can be performed utilizing traditional indices like B+-trees or R-trees. The drawback of this approach is that traditional indices do not achieve good temporal data clustering. Better clustering is achieved by temporal indices through record copying. Nevertheless, record copies can greatly affect the correctness and effectiveness of join performance.