A Database Machine for Very Large Relational Databases

This paper presents an architectural design for a back-end database machine (DBM) suitable for supporting multiple-user, on-line, very large relational database systems (systems with capacities larger than 1010 bytes). This machine is called the Michigan Relational Database Machine (MIRDM). In designing such a machine a structured approach is followed. First, the DBM´s proposed so far are reviewed using a novel classification scheme. Second, this review, the very large relational database system requirements, and the restrictions imposed by the current and near-future state of technology are used to formulate a set of design guidelines. Consequently, an architecture for a cost-effective DBM that meets this set of guidelines is synthesized. Some of the algorithms developed for implementing the primitives of the new machine are also presented. Finally, we highlight some of the important differences between the new DBM and some of the previously proposed ones.