Connection Machine

from Wikipedia, the free encyclopedia
A CM-2 in the Computer Museum in San Jose.

The Connection Machine was a series of parallel computers that was manufactured from 1983 to 1991 by the American company Thinking Machines .

Emergence

The Connection Machine concept came from Danny Hillis of the Massachusetts Institute of Technology (MIT). Together with Sheryl Handler, he founded Thinking Machines in 1983 ; the development of the computers was financed by venture capital and the American Department of Defense ( DARPA ) and later supported by the High Performance Computing and Communication (HPCC) program. In August 1993, Thinking Machines had to file for Chapter 11 bankruptcy.

Construction and variants

The first connection machine was a massively parallel system with up to 65536 1-bit processors . Each processor could communicate directly with 20 others via a Hypercube interconnection network. The processor nodes had their own main memory and initially worked according to the SIMD principle.

The CM-1 (1983) was primarily designed to solve problems in the field of artificial intelligence . That's why * Lisp (also Star Lisp , a parallel extension of Common Lisp ) was used for programming. With the CM-2 (1987), the Connection Machine also became interesting for numerical processes. Each 32 processor nodes shared a coprocessor (Weitek 3132), a so-called SPRINT chip was used as an interface to this , which u. a. had the ability to generate a 32-bit number from the 32 1-bit processor nodes. Together, the processor nodes achieved an output of up to nine (theoretically 20) G FLOP (for comparison, a normal PC with a Pentium 4 processor at a clock frequency of three gigahertz can achieve around six GFLOPS, according to IBM). The CM-2a was a smaller version with 4096 or 8192 processors, the CM-200 was a further development of the CM-2. In addition, with the introduction of the CM-2, it could also be programmed via C * (a parallel extension of C ) and CM Fortran .

A change in the computer architecture towards MIMD then took place in 1991 with the CM-5. It consisted of a Fat Tree interconnect network of SPARC processors. In the CM-5E, the SPARC-V7 processors were eventually replaced by SuperSPARC processors.

While connection machines were used via front-end computers ( Symbolics , VAX and later also SPARCstations ), CM-5 was the first connection machine that also ran its own operating system , called CMost , which was based on SunOS, but was only based on Control processors were used, which provided users with logins and network services. A small microkernel operating system ran on the individual processor elements, which was loaded from a ROM at startup and provided basic functions for accepting and executing jobs.

design

The Connection Machine also stood out for its design. The case was a large block, mostly cube-shaped. On the front there were groups of red light emitting diodes ; there was a light emitting diode for each processor. The blinking of the LEDs signaled the activity of the individual processor nodes. A CM-5 can be seen in the movie Jurassic Park (1993).

literature

  • W. Daniel Hillis: The Connection Machine. 1985 (MIT Press Series in Artificial Intelligence) ISBN 0-262-08157-1

Individual evidence

  1. ^ Arthur Trew and Greg Wilson: Past, Present, Parallel: A Survey of Available Parallel Computing Systems. 1991, pp. 38-40 ISBN 0-387-19664-1
  2. ↑ Connection Machine CM-5 Technical Summary, 1993, pp. 37-43

Web links