The dataflow database machine is designed to be a highly parallel database machine with many processing elements and many small disks. Since I/O bandwidth is a necessary but not sufficient condition to guarantee better performance, the major theme of the paper is how to usefully capture this potential increase in bandwidth. The authors address this problem first in the form of a computational model, second in the model’s realization as an architecture, and last in such issues as concurrency control, recovery from hardware and software failures, and the interleaving of nonserializable transactions. The most interesting aspect of this approach is the active graph model, the AGM of the title, which is both the dataflow database machine’s conceptual organization and its method of executing transactions. It achieves its design goals of having no visible parallelism in queries, distributed control and storage, and the ability to process many requests concurrently. In the AGM’s conceptual organization, each node is an entity or relationship in the entity-relationship (ER) model, and related elements of an entity, or relationships, are connected by bidirectional arcs. Each node is an autonomous processing element, giving the AGM its dataflow aspect.
The AGM loses some of its elegance when mapped onto an architecture. Among the other hazards of reality, the system can never have enough processing elements for one per graph node, and the nodes can never be fully interconnected. The hardware implementation section comprehensively describes the major architectural characteristics of the dataflow database machine. The methods of achieving data integrity and failure recovery, which are especially hard to build in a dataflow machine and especially important to have in a database machine, have a workable design. Although performance is discussed for the individual disk storage components, the authors provide no discussion or prediction of complete system performance. Otherwise this paper is well written and presents an interesting, significant piece of work. I look forward to reading the authors’ forthcoming paper in which they promise to present performance results from their recently built prototype.