The Connection Machine describes a fundamentally different kind of computer. It offers a preview of a computer that Daniel Hillis and others are now developing to perform tasks that no conventional sequential machine can solve in a reasonable time.
Today’s conventional computers are characterized by sequential serial operations proceeding back and forth between a single core processor and a separate large-scale memory. The Connection Machine, by contrast, is a general, universal parallel computer, of which thousands or millions of independent simple processes – each integrated with its own small but sufficient memory – operate concurrently on different aspects or segments of the same overall problem.
In the Connection Machine all these processes can be connected with each other in any desired pattern. The programming strategy calls for arranging the processing elements in a way that matches as closely as possible the natural structure of the problem. Moreover, these patterns of connection can be initially established by the program, or they can be reconfigured dynamically as the computation proceeds and conditions change.
The Connection Machine considers how computer architecture can be used to solve problems and how the architecture can be implemented using available technology. It deals with both the programming and the building of computers, with both software and hardware and their ultimate integration in the design process.
The book presents a programming language based on Lisp and discusses an actual prototype, various active data structures, storage allocations, Connection Machine physics, and the likely evolution of the science of computation.
W. Daniel Hillis is a founder of Thinking Machines Corporation, where he is engaged in building Connection Machines as a significant step toward real thinking machines.
The Connection Machine is seventeenth in the MIT Press Artificial Intelligence series, edited by Patrick Henry Winston and Michael Brady.