Theory of Adaptive Mechanisms. Part V. A New Approach Toward Parallel Processing

摘要

Based on the same logic circuit delay and main memory cycle, the performance of a computer system can be improved by better system organization and more efficient logic design. The general approach is by means of more parallel data paths and many processors operating simultaneously. A computer may be so designed that a sequence of instructions of a program can be executed concurrently whenever possible. It is called a confluent system. Another approach employs many processors connected as a matrix. These processors have a control unit which controls the instruction and data flow. It is very fast if the problem can be formalized in terms of matrices. A third approach is a multicomputer system in which several computers are connected through their common I/O devices and/or special data channels. A study of these systems leads to a new system organization. It has a variable configuration and can be used as either a multiprocessor system or a multicomputer system. It contains many program executors operating under a hardware supervisor. The supervisor itself is a program executor which can be assigned by the operator from one of a few special program executors. The unassigned units can be used as ordinary program executors. The basic difference of the proposed system from the other systems is that the program executors are very loosely coupled. (Author)