It has long been recognized that the major obstacle to performing I/O-intensive computations such as database processing with the traditional von Neumann architecture is the processor-memory bottleneck. Von Neumann computers are best suited for computations with complex control structures, no parallelism, and a small amount of data. Data retrieval, on the other hand, involves simple operations, large volumes of data, and much potential for parallelism. This observation provided the motivation for the design and development of database machines.
This paper was first published in 1986 in the proceedings of the Twelfth Very Large Database Conference. It reports on a relational database machine called GAMMA and was written by a group that has been active in database machine research for over a decade. One merit of this paper is that the authors compare and contrast GAMMA with their previous attempt (which resulted in another system called DIRECT). The basic building blocks of GAMMA are disk drives. Each unit has its own processor; the processors are interconnected via a Cambridge Ring network, and the control overhead is reduced by using dataflow processing techniques. The prototype reported in this paper involves 20 VAX 11/750 computers connected by an 80 megabit/second ring. The paper describes various components of the machine, explains the query execution schemes and the associated algorithms, and compares performance results under different schemes. Overall, it is interesting and well-written.