The Bulk Index Join: A Generic Approach to Processing Non-Equijoins

Efficient join algorithms have been developed for processing different types of non-equijoins like spatial join, band join, temporal join or similarity join. Each of these previously proposed join algorithms is tailor-cut for a specific type of join, and a generalization of these algorithms to other join types is not obvious. We present an efficient algorithm called bulk index join that can be easily applied to a broad class of non-equijoins. Similar to the well-known hash join algorithms, the bulk index join performs in two phases. In the build-phase, an appropriate index structure is created that serves as a partitioning function on the first relation. In the probing-phase, the records of the second relation are probed against the first relation by using the index structure of the build-phase. In order to support both phases efficiently, we adopt a technique recently proposed for bulk loading index structures. We show that this technique can also be exploited for probing the tuples of the second relation in bulk. Similar to the generic bulk loading approach, only a predefined set of routines of the index structure is used for implementing our join algorithm. This set is generally available in tree-based index structures. The so-called band join serves as an example in this paper. We first discuss in detail how to apply our generic approach to the band join. Thereafter, we present a worst-case analysis and experimental results. Moreover, we show in our experiments that the well-known index nested loops join can benefit from performing queries in bulk as it is proposed for the probing-phase of the bulk index join.