A database machine based on the data distribution approach

Various VLSI circuits, each of which realizes a specific database operation, have been studied; and a VLSI database machine can be created by a collection of these circuits. Such a method is called the function distribution approach. The problems of this approach are that (1) the data transmission cost is very high and (2) some circuits become very slow when the data size exceeds the maximum size handled by the circuits. Since database systems handle a large number of data, we need to develop another approach that costs less for data transmission and has expandability, Because most database operations can be divided into operations on subsets of data, this paper proposes the data distribution approach. In this approach a subset of data is stored in a functional storage circuit, and each circuit can realize most database operations. The whole system can be viewed as a file system having functions for database operations. Compared with conventional file systems, the system has the following advantages: (1) frequent rebalancing is not required, and (2) parallel processing of database operations is realized. Three methods to realize functional storage circuits are described. Selection is made by cost, performance, and available VLSI technology. An organization of such circuits with efficient database processing is discussed in detail; it will be realized by technology in the near future.