Lisp machine
A Lisp machine (or LispM for short ) is a computer whose processor is optimized for the execution of LISP programs and whose operating system and typical applications are written in LISP. Lisp machines offered a convenient integrated development environment . Some Lisp machines were also used without a development environment for the use of Lisp applications. Lisp machines were designed in the 1970s and 1980s to better handle artificial intelligence (AI) tasks. Some machines were also used for animation . The number of Lisp machines used was very small. There are estimates between 5000 and 7000 pieces.
Despite this small number some concepts of today's computer systems on Lisp machines were tested and first used productively as color graphics, window systems , computer mice , computer networks , hypertext , incremental compilation and the concept of single-user workstations .
Because of the great advances in processor technology in the 1990s, which also made sufficiently fast general microprocessors possible, and the AI winter , the Lisp machine manufacturers plunged into a crisis and the production of Lisp machines ended.
Technical idea behind the Lisp machines
During the 1960s and 1970s, the Lisp programming language was mostly used for research projects on artificial intelligence . The computers available were optimized to process programs in languages such as Assembler or Fortran with as little memory consumption and computing time as possible. The execution of Lisp programs, on the other hand, required considerable resources for the time. One reason for this was the increasingly complex Lisp programs and the large amounts of data that were processed in AI applications.
Lisp uses dynamic typing and dynamic memory management ( garbage collection ). The Lisp machines often offer generic operations. The machine operation + accepts any number types at runtime. The processor determines the types of the arguments, checks them for applicability, carries out conversions, selects the appropriate addition and then applies it. For this purpose, the data words were provided with type information (tagged) . The type check could take place in parallel and was much faster than a software implementation. Typical word lengths for Lisp machines are 32 bits (e.g. TI Explorer Microprocessor), 36 bits (Symbolics 3600) or 40 bits ( Symbolics Ivory ). With a word length of 36 bits, data with 32 bits and tags with 4 bits also fit into one data word.
In addition, virtual memory was introduced and garbage collection was supported by the hardware. In commercial Lisp machines, entire Lisp functions were implemented in hardware.
In order to offer users as much computing power as possible, Lisp machines were designed as single-user computers (with support for bitmap screens, keyboard, mouse, network interface, hard drives , tape drives and various expansion slots). This was unusual at the time when mainframes were used as a multi-user system via terminals . In order to enable several users to work together, Lisp machines were designed with the ability to form computer networks (initially Chaosnet , later also Ethernet ), which was also unusual for the time.
Lisp machines were also leading in the field of hypertext . The documentation system of the Lisp machine manufacturer Symbolics won several awards.
Not only Lisp programs run on Lisp machines. There are also compilers for, for example, C , Pascal , Fortran , Ada and Prolog . These compilers were mostly also written in Lisp and can be used just as interactively as the Lisp compiler.
history
Started at MIT AI Lab and Xerox
In 1973 Richard Greenblatt and Tom Knight began developing a prototype for a machine that would run Lisp code in an optimized way. The first machine that Knight wrote his master's thesis on was called CONS machine (after the Lisp function cons
) and was completed in 1976. The CONS machine had a 24-bit architecture and still required a PDP-10 for operation. After the machine was presented at a 1978 Artificial Intelligence conference, DARPA began funding the project and companies expressed interest in purchasing a Lisp machine. This led to the development of the CADR machine (based on the Lisp function cadr
), of which 25 pieces were produced. The great interest in Lisp machines led to the creation of a company for marketing.
In parallel to the development at MIT , BBN Technologies developed its own Lisp machine (Jericho) , which, however, was never marketed. The disappointed team was then poached by Xerox and developed a Lisp machine called Dolphin at Xerox PARC in 1979 . The Xerox Lisp machines were based on InterLisp , as opposed to the MIT machines, which were based on Maclisp .
Commercialization: Symbolics Inc., Lisp Machines Inc.
In 1979 there was a dispute between Russell Noftsker and Greenblatt over the company's business model. Noftsker wanted to build a traditional company, while Greenblatt wanted a business model that was compatible with the hacking ethics of the MIT AI Lab and that would forego venture capital . Since Noftsker, who had left the AI Lab in 1973 to work in the private sector, already had experience in the commercial world and there was also a dispute between Greenblatt and some employees at the AI Lab for other reasons, Noftsker managed many employees to winning among others Thomas Knight, for his plans and he founded Symbolics Inc . Greenblatt initially remained passive and was very upset with Noftsker. Control Data Corporation (CDC), however, showed great interest in acquiring an MIT CADR machine; Alexander Jacobson, a consultant from CDC, therefore persuaded Greenblatt to finally found their own company, Lisp Machines Inc. (LMI). In 1980/1981 Symbolics launched the LM-2 , which was a "repackaged" MIT-CADR machine. LMI also released an MIT-CADR machine, the LMI-CADR machine.
The competition between LMI and Symbolics meant that the employees of both companies had to leave the AI Lab. Only Richard Stallman and Marvin Minsky stayed behind. In addition, LMI and Symbolics had licensed their technology and software from MIT and granted MIT a right to use their changes, but Symbolics refused MIT to integrate the changes into the original prototype and software so that LMI could not use them. This angered Stallman, who became a free software advocate . Stallman used the access at MIT to the Lisp machines to reconstruct the changes and to provide LMI. However, LMI also kept its own changes to the Lisp machine closed.
LMI licensed their Lisp machines to Texas Instruments , which produced machines based on the LMI Lambda with Explorer I / II.
End of the Lisp machines
Towards the end of the 1980s and beginning of the 1990s, the already small market for Lisp machines collapsed. LMI was already insolvent in 1986 and an attempt to revive the company as GigaMos Systems failed due to legal problems of the investor. Xerox stopped developing other Lisp machines relatively early on.
There are many reasons for the collapse. For one thing, the market was very small. Speculations assume between 5000 and 7000 machines. This ensured that the manufacturers could invest less money in the technical further development of the Lisp machines, while the manufacturers of conventional computers developed ever better processes and Lisp machines soon caught up with speed and even overtook them. Firms like Lucid Inc. and Franz Inc. started selling Lisp environments for microcomputers . The porting of the Symbolics operating system software Genera from 1992 to a DEC Tru64 UNIX / Alpha system was three times as fast as the fastest Lisp machine.
In addition, the exaggerated expectations of artificial intelligence were not fulfilled, which is why the funds for many AI research projects were cut (the so-called AI winter ). The massive cut in funding for the SDI project (also known as the Star Wars project) hit the market hard. Many AI research projects (especially in the area of expert systems ) were funded by SDI funds. The most important market for Lisp machines collapsed.
Manufacturer of Lisp machines
- Symbolics
- Lisp Machines Inc. (LMI)
- Integrated Inference Machines (IIM)
- Texas Instruments
- Fujitsu
- Xerox
Lisp machines
Publishing year | Surname | Specialty |
---|---|---|
1975/1976 | WITH CONS | first Lisp machine; 24 bit architecture |
1977/1978 | WITH CADR | |
1979 | Xerox 1100 (Dolphin) | based on the Xerox Alto |
1980/1981 | Symbolics LM-2 | "Repackaged" WITH CADR |
1980/1981 | LMI CADR | "Repackaged" WITH CADR |
1981 | Xerox 1108 (Dandelion) | based on the Xerox Star |
1982 | Symbolics 3600 | with 36-bit data word and 28-bit address space; including Macsyma and prologue |
1982 | Xerox 1109 (Dandetiger) | identical to the Xerox 1108, but larger memory expansion |
1983 | LMI Lambda | |
1983 | Texas Instruments Explorer I. | LMI Lambda produced under license |
1983 | Xerox 1132 (Dorado) | |
1984 | Symbolics 3650 | |
1985 | Xerox 1185/1186 (Dove / Daybreak) | 1185 only with runtime environment , 1186 with complete programming environment |
1985 | Texas Instruments Lisp chip | integrated microprocessor; 32 bit |
1986 | Fujitsu FACOM alpha | Lisp and Prolog coprocessor for Fujitsu mainframes ; only Japanese Lisp machine |
1986 | LMI K-Machine | completely new hardware design; integrated microprocessor; could not be completed due to LMI's insolvency |
1986 | Symbolics 3620 | |
1986 | IIM | IIM produced some prototypes |
1987 | Symbolics Ivory | integrated microprocessor; 40 bits + 8 bits ECC; Basis for several Lisp machines |
198? | Symbolics XL400 | Ivory-based workstation with VME bus |
1987 | Texas Instruments Explorer II | is based on the Lisp chip from TI |
1988 | Symbolics MacIvory I. | Nubus plug-in card for Apple Macintosh computers, Ivory-based |
1988 | Symbolics XL400 | Ivory-based workstation with VME bus |
1988 | Texas Instruments MicroExplorer | Nubus plug-in card for Apple Macintosh computers, based on the Lisp chip from TI |
1989 | Symbolics MacIvory II | Nubus plug-in card for Apple Macintosh computers, Ivory-based |
1989 | Symbolics UX400 | Ivory-based VME bus coprocessor board for SUN computers |
1990 | Symbolics XL1200 | Ivory-based workstation with VME bus |
1990 | Symbolics UX1200 | Ivory-based VME bus coprocessor board for SUN computers |
1991 | Symbolics MacIvory III | Nubus plug-in card for Apple Macintosh computers, Ivory-based |
1992 | Symbolics XL1201 | Ivory-based desktop workstation with VME bus |
1992 | Symbolics NXP1000 | Ivory-based Lisp machine without its own display |
1993 | Symbolics Open Genera | virtual Lisp machine for Tru64 UNIX on DEC Alpha |
literature
- Paul Graham : Anatomy of a Lisp Machine. In: AI Expert. Vol. 3, No. 12, December 1988, ISSN 0888-3785 , pp. 26-32.
- Andrew R. Pleszkun, Matthew J. Thazhuthaveetil: The Architecture of Lisp Machines. In .: IEEE Computer. Vol. 20, No. 3, March 1987, pp. 35-44.
- Charles L. Ditzel, Douglas Schuler, Virginia Thomas: A Lisp Machine Profile. Symbolics 3650. In: AI Expert. Vol. 2, No. 1, January 1987, pp. 69-73
- Peter M. Kogge: The Architecture of Symbolic Computers. (= Mc-Graw-Hill series in supercomputing and parallel processing ) McGraw-Hill, New York a. a. 1991, ISBN 0-07-035596-7 .
Web links
- Lisp machines ( Memento from July 9, 2015 in the Internet Archive )
- Lisp Machine Manual (English)
- Lispm FAQ and Oral History (English)
- Lisp machine emulators (MIT CADR, Symbolics, TI Explorer, ...) and original MIT CADR source code (English)
Individual evidence
- ↑ Lispm FAQ and Oral History ( Memento of the original dated November 8, 2007 in the Internet Archive ) Info: The archive link was inserted automatically and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice. on andromeda.com
- ^ Symbolics Inc: Symbolics Technical Summary ( Memento November 1, 2012 in the Internet Archive ), 1985
- ↑ a b c Alvin Graylin, Kari Anne Hoier Kjolaas, Jonathan Loflin, Jimmie D. Walker: Symbolics, Inc .: A failure of heterogeneous engineering. ( Memento from May 9, 2013 in the Internet Archive )
- ^ A b c Steven Levy: Hackers: Heroes of the Computer Revolution. New York 1984, ISBN 0-385-19195-2
- ↑ Computing Facilities for AI: A Survey of Present and Near-Future Options, AI Magazine Volume 2 Number 1
- ↑ A Brief History of Lisp Machines ( Memento of the original from August 13, 2006 in the Internet Archive ) Info: The archive link was inserted automatically and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice. on andromeda.com
- ^ A b My Lisp Experiences and the Development of GNU Emacs on gnu.org
- ^ A b Dan Weinreb: Rebuttal to Stallman's Story About The Formation of Symbolics and LMI ( Memento of December 13, 2007 in the Internet Archive ) , November 11, 2007
- ^ Lisp Machine Inc. K-machine on tunes.org
- ↑ OpenGenera Benchmarks ( Memento from April 20, 2014 in the Internet Archive ) on blog.b9.com
- ↑ The Lisp Machine: Noble Experiment Or Fabulous Failure? on withy.org
- ^ Martin Fransman: The market and beyond: information technology in Japan . Cambridge University Press, ISBN 978-0-521-43525-3 ( Limited preview for Google Book Search).
- ↑ Evaluation of the FACOM ALPHA Lisp machine on acm.org