Taking Constraints out of Constraint Databases

By providing a finite representation for data with infinite semantics, the Constraint Database approach is particularly appropriate for querying spatiotemporal data. Since the introduction of CDBs in the early 1990’s, several CDB prototypes have been implemented; however, CDBs have yet to prove their commercial viability. In this paper, we rethink the main contribution behind the Constraint Database (CDB) framework; it’s not about constraints! Rather, the main contribution of CDBs is in providing an additional layer to the database architecture, by splitting the logical into abstract and concrete layers. The abstract layer, representing an infinite relational extent of the spatiotemporal data, is the one over which CDB queries are written. The finite data representation is in the concrete layer; traditionally, constraints are used for this representation. By contrast, in current GIS and spatial databases, the representation is based on geometrical structure rather than constraints. We propose a single three-tier architecture that combines these ap-proaches. Its top abstract layer represents the infinite relational extent of the data, but the concrete layer admits both constraint-based and geometric data representations. The resulting constraint-backed systems, while preserving the advantages of CDBs, are more practical than pure constraint databases. By admitting more flexible CDB architectures, where data is not necessarily represented with constraints, significant progress can be made towards producing commercially successful CDB systems.