Indexing on Multivalued Attributes: Partial-Order Trees vs. Bitmaps

Abstract

The paper presents two methods of indexing on attributes whose values are sets of primitive values. Attributes of such types are common in object-oriented and object-relational database systems and go beyond the first-normal form paradigm of relational data model. The paper introduces the concept of set-index and presents adequate index structures. The first index structure is based on a partial-order tree that is a tree derived from a partial-order relation imposed on sets with inclusion operator. For this index organisation, evaluating set predicates consists in traversal a multiway tree in a systematic fashion. The second structure is based on an array of bitmaps. For this structure, set predicates can be evaluated by retrieving appropriate rows from an array of bitmaps, depending on the index variant. Two variants of the latter structure are proposed: with fixed-sized rows and with variable-size rows. The index structures have been analysed and compared in context of time and space efficiency, flexibility in dynamic environments and other factors important for performing set predicates referring to whole sets.