An associative search language for data management

Recent years have witnessed a widespread and intensive effort to develop systems to store, maintain and access data bases of varied size. Such systems are referred to as DBMS-Data Base Management Systems. In different areas, such as artificial intelligence, management information systems, military and corporate logistics and medical diagnosis, a wide variety of DBMS exist. All these systems have generally been implemented on conventional computers, which are based on the von Neumann design. In this design, operations will be performed on the information in the memory by means of their addresses. Because of the size of typical data bases and costs of memory, we cannot hold all information in the main memory and swapping converts the search problem to a transportation problem. Present-day systems have to transfer large sets of data from their mass storage to the CPU, where simple compare-functions are performed in order to separate relevant data from irrelevant data. The transfer channels with their limited capacity form the main bottleneck of this system and as a result, great efforts have been made to reduce the necessary data flow by means of sophisticated software systems and additional redundancy such as index tables and inverted files. By these techniques, address of information will be obtained from a directory. Although directory partially solved the bottleneck problem, it nevertheless created some problems. The directory should logically be kept in the main memory. If we are dealing with large data bases, naturally we are dealing also with large directories, and large directories occupy a large portion of the memory. Also, the use of directories will create some complexity in the search, update and delete algorithms. Conventional computers are all based on numerical operations. The necessity of designing new hardware based on non-numerical operations has been discussed in detail by one of the authors.1 In contrast, use of associative or content addressable memories and hardware design based on non-numerical operations as well as numerical operations causes information stored at unknown locations to be processed efficiently on the basis of some knowledge of its content. In developing new architectures for future machines some of the most important trends in hardware and software technology must be brought into focus. On the hardware side,