Chaosnet wiki

The global Chaosnet site

User Tools

Site Tools

Trace:
tr173

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Next revision		Previous revision
tr173 [2020-12-16 17:47] – created victortr173 [2021-05-17 17:44] (current) – [6. User's impressions of the Chaosnet/SUPDUP MIT environment] victor
Line 1: Line 1:
 ====== LOCAL NETWORKING AT THE M.I.T. COMPUTER LABORATORIES: A User's View ====== ====== LOCAL NETWORKING AT THE M.I.T. COMPUTER LABORATORIES: A User's View ======
 Marek W. Lugowski  Marek W. Lugowski 
 +
 Indiana University Computer Science Department  Indiana University Computer Science Department 
 +
 Technical Report #173  Technical Report #173 
 +
 November 1984 - June 1985 November 1984 - June 1985
  
Line 14: Line 17:
 Local base-band area networks, time-sharing terminal, file-transfer protocols, design of distributed computing environments, lisp machine, history of artificial intelligence, office systems Local base-band area networks, time-sharing terminal, file-transfer protocols, design of distributed computing environments, lisp machine, history of artificial intelligence, office systems
  
-===== Environment =====+===== - Environment =====
 The computing environment described in this paper is an important subset of MIT's total computing environment, and it consists of The Artificial Intelligence Laboratory (henceforth, AI Lab), The Laboratory for Computer Science (LCS), the largest mainframes in the Department of Electrical Engineering and Computer Science (EECS), and Symbolics, the nearby upstart company launched by the one-time MIT "moby wizards" to develop and manufacture Lisp Machines. Lisp Machines were invented at MIT by, among others, the people who are now at Symbolics, and one could say that Symbolics and MIT enjoy The computing environment described in this paper is an important subset of MIT's total computing environment, and it consists of The Artificial Intelligence Laboratory (henceforth, AI Lab), The Laboratory for Computer Science (LCS), the largest mainframes in the Department of Electrical Engineering and Computer Science (EECS), and Symbolics, the nearby upstart company launched by the one-time MIT "moby wizards" to develop and manufacture Lisp Machines. Lisp Machines were invented at MIT by, among others, the people who are now at Symbolics, and one could say that Symbolics and MIT enjoy
 a close symbiotic relationship, to the point of sharing the same local area network. This uncommon--in the academia--intermingling of the university with the corporate is characteristic of MIT. The LCS and the AI Lab, for example, share their building with Polaroid, and a third party still owns and operates the premises. a close symbiotic relationship, to the point of sharing the same local area network. This uncommon--in the academia--intermingling of the university with the corporate is characteristic of MIT. The LCS and the AI Lab, for example, share their building with Polaroid, and a third party still owns and operates the premises.
Line 26: Line 29:
 is connected to 545 Technology Square (the MIT computer science labs) via a roof-mounted microwave link. is connected to 545 Technology Square (the MIT computer science labs) via a roof-mounted microwave link.
  
-===== Inventing equipment ahead of the marketplace =====+===== - Inventing equipment ahead of the marketplace =====
  
 A good deal of the above environment's hodge-podge character owes itself to the MIT AI Lab's penchant for designing and constructing equipment that only later becomes commercially available. The net result of this activity is, of course, a great contribution to computer technology in general, but also a chaotic, home-made look of the equipment in the AI Lab's machine room. In fact, quite often there may be around only one or two people qualified to maintain a A good deal of the above environment's hodge-podge character owes itself to the MIT AI Lab's penchant for designing and constructing equipment that only later becomes commercially available. The net result of this activity is, of course, a great contribution to computer technology in general, but also a chaotic, home-made look of the equipment in the AI Lab's machine room. In fact, quite often there may be around only one or two people qualified to maintain a
 particularly obscure device, and certainly no outside repairman would be willing to mess with this hardware. The home-made core memories, the home-made lisp machines, and the home-made peripheral devices, as well as the bridges connecting Chaos links, all fit in this category. particularly obscure device, and certainly no outside repairman would be willing to mess with this hardware. The home-made core memories, the home-made lisp machines, and the home-made peripheral devices, as well as the bridges connecting Chaos links, all fit in this category.
  
-===== Problems peculiar to the MIT computer labs =====+===== - Problems peculiar to the MIT computer labs =====
  
 The labs have a long tradition of using the DEC PDP-10 computers. Each lab has a huge DecSystem-20 and a couple smaller KI-10 systems, some almost on the way out. The hackers at the labs have been historically fond of these machines, and a lot of them refuse to hack on anything else but those (and, of course, the lisp machines). The labs have a long tradition of using the DEC PDP-10 computers. Each lab has a huge DecSystem-20 and a couple smaller KI-10 systems, some almost on the way out. The hackers at the labs have been historically fond of these machines, and a lot of them refuse to hack on anything else but those (and, of course, the lisp machines).
Line 50: Line 53:
 Despite these differences, the two labs are condemned to share the same facilities for output, the same machine room, and--of course--the same aggregate of local area networks. The joint computer labs at MIT may well be the most exasperating territory for a systems designer and a systems programmer to work at. Despite these differences, the two labs are condemned to share the same facilities for output, the same machine room, and--of course--the same aggregate of local area networks. The joint computer labs at MIT may well be the most exasperating territory for a systems designer and a systems programmer to work at.
  
-===== Chaosnet, MIT's own local area network architecture =====+===== - Chaosnet, MIT's own local area network architecture =====
  
-==== INTRODUCTION ====+==== - INTRODUCTION ====
 Before the Ethernet was commercially available, the people at the MIT computer science laboratories decided to develop and install their own local area network. The need was motivated by the parallel effort at developing the lisp machine. Lisp machine, conceptually a multiprocessor, was to comprise a set of dedicated CPUs, memories, and swapping disks, but with a central file system. The dedicated CPU, disk, and memory were provided in order to assure a constant level of high quality performance. The shared file system was designed to facilitate centralized back-up and maintenance, as well as to provide mail interaction, the ability to send messages interactively, and the sharing of programs. The Chaosnet was to be the file-transfer medium and the communications bus of this system. Before the Ethernet was commercially available, the people at the MIT computer science laboratories decided to develop and install their own local area network. The need was motivated by the parallel effort at developing the lisp machine. Lisp machine, conceptually a multiprocessor, was to comprise a set of dedicated CPUs, memories, and swapping disks, but with a central file system. The dedicated CPU, disk, and memory were provided in order to assure a constant level of high quality performance. The shared file system was designed to facilitate centralized back-up and maintenance, as well as to provide mail interaction, the ability to send messages interactively, and the sharing of programs. The Chaosnet was to be the file-transfer medium and the communications bus of this system.
  
Line 66: Line 69:
 While predating the commercial availability of the Ethernet, the Chaosnet project heavily borrowed on the research already done at Xerox PARC, as well on the concepts developed in TCP and the Arpanet. Nowadays, a lot of the hardware on which the Chaosnet runs is the very same thing supplied by the Ethernet vendor. While predating the commercial availability of the Ethernet, the Chaosnet project heavily borrowed on the research already done at Xerox PARC, as well on the concepts developed in TCP and the Arpanet. Nowadays, a lot of the hardware on which the Chaosnet runs is the very same thing supplied by the Ethernet vendor.
  
-==== HARDWARE ====+==== - HARDWARE ====
  
 The physical medium of the Chaosnet is the 75-ohm TV cable. The physical medium of the Chaosnet is the 75-ohm TV cable.
Line 90: Line 93:
 The interface is a wire-wrap board containing some 120 TTL chips. It implements the network's hardware protocols, buffers the packets going both ways, handles the error-checking, and interrupts the host when a packet comes in or out of the buffers. The interface provides packet synchronization so that the host may generate packets asynchronously. There are interfaces for lisp machines, LSI-11's, and the DEC Unibus. The Unibus covers all of the other interfacing needs if something doesn't have a Unibus, it is connected to a bridge. The interface is a wire-wrap board containing some 120 TTL chips. It implements the network's hardware protocols, buffers the packets going both ways, handles the error-checking, and interrupts the host when a packet comes in or out of the buffers. The interface provides packet synchronization so that the host may generate packets asynchronously. There are interfaces for lisp machines, LSI-11's, and the DEC Unibus. The Unibus covers all of the other interfacing needs if something doesn't have a Unibus, it is connected to a bridge.
  
-==== HARDWARE PROTOCOLS ====+==== - HARDWARE PROTOCOLS ====
  
  These protocols deliver packets from node to node on the same cable. They provide a decent probability of successful delivery and packet integrity (i.e., if a packet is compromised, say so or discard it).  These protocols deliver packets from node to node on the same cable. They provide a decent probability of successful delivery and packet integrity (i.e., if a packet is compromised, say so or discard it).
Line 122: Line 125:
 nodes will not collide again on the retransmission in a periodic lock. nodes will not collide again on the retransmission in a periodic lock.
  
-==== OVERVIEW OF SOFTWARE PROTOCOLS ====+==== - OVERVIEW OF SOFTWARE PROTOCOLS ====
  
  These protocols arrange for high-speed interchanges between hosts, regardless of location (hosts may reside on different cable links), and without undetected transmission errors. The original design goals called for the speed of file transfers to be comparable to that of inexpensive tape drives, 3 K characters per second,  These protocols arrange for high-speed interchanges between hosts, regardless of location (hosts may reside on different cable links), and without undetected transmission errors. The original design goals called for the speed of file transfers to be comparable to that of inexpensive tape drives, 3 K characters per second,
Line 166: Line 169:
 now popularized through HDLC protocols. An acknowledgment obviates the need for sending receipts. Each higher-level protocol has a pre-adjusted window size, and although there is a facility for communicating dynamic adjustment of windows, this is never done in practice, to avoid introducing complexity into the protocols. now popularized through HDLC protocols. An acknowledgment obviates the need for sending receipts. Each higher-level protocol has a pre-adjusted window size, and although there is a facility for communicating dynamic adjustment of windows, this is never done in practice, to avoid introducing complexity into the protocols.
  
-==== HIGHER-LEVEL PROTOCOLS ====+==== - HIGHER-LEVEL PROTOCOLS ====
  
  
Line 174: Line 177:
 one of the PDP-10s which act as Arpanet gateways, TIME provides the number of seconds elapsed since January 1, 1900 Greenwich Mean Time as a 32-bit number. Lisp machines, which do not have hardware calendar clocks, use TIME to find out the date and time when they first come up for service. The SUPDUP protocol is discussed later in this paper; it is a ment of TELNET. one of the PDP-10s which act as Arpanet gateways, TIME provides the number of seconds elapsed since January 1, 1900 Greenwich Mean Time as a 32-bit number. Lisp machines, which do not have hardware calendar clocks, use TIME to find out the date and time when they first come up for service. The SUPDUP protocol is discussed later in this paper; it is a ment of TELNET.
  
-==== FOREIGN PROTOCOLS IN Chaosnet ====+==== - FOREIGN PROTOCOLS IN Chaosnet ====
  
 Any foreign protocol based on the idea of a full-duplex stream (or 2 half-duplex streams) of 8-bit bytes can be simply incorporated into Chaosnet using the Chaosnet connection mechanism instead of whatever stream protocol the foreign protocol was originally using. This was the case with Arpanet's TELNET. Any foreign protocol based on the idea of a full-duplex stream (or 2 half-duplex streams) of 8-bit bytes can be simply incorporated into Chaosnet using the Chaosnet connection mechanism instead of whatever stream protocol the foreign protocol was originally using. This was the case with Arpanet's TELNET.
Line 180: Line 183:
 There is a facility known as the Chaosnet foreign-protocol protocol which allows alien packets to be transmitted through Chaosnet or allows two Chaosnet hosts to speak in non-Chaos manner to each other. Occasionally non-stream I/O devices, such as graphics tablets, need to be connected to the network and this facility makes it possible. There is a facility known as the Chaosnet foreign-protocol protocol which allows alien packets to be transmitted through Chaosnet or allows two Chaosnet hosts to speak in non-Chaos manner to each other. Occasionally non-stream I/O devices, such as graphics tablets, need to be connected to the network and this facility makes it possible.
  
-===== SUPDUP -- Chaosnet's remote login facility =====+===== - SUPDUP -- Chaosnet's remote login facility =====
  
  
  The SUPDUP's main claim to fame is that it allows for remote logins in a manner of terminal-independent output. This way, only the local system needs to know how to handle the user's terminal. SUPDUP also has a built-in graphics interface as well as local assistance for  The SUPDUP's main claim to fame is that it allows for remote logins in a manner of terminal-independent output. This way, only the local system needs to know how to handle the user's terminal. SUPDUP also has a built-in graphics interface as well as local assistance for
-the remote text editors (i.e., text editors used on the remote machine by the local user). SUPDUP means to be a superior meat of the Arpanet's TELNET.+the remote text editors (i.e., text editors used on the remote machine by the local user). SUPDUP means to be a superior ment of the Arpanet's TELNET. 
 Both TELNET and SUPDUP define a virtual terminal, but TELNET's  Both TELNET and SUPDUP define a virtual terminal, but TELNET's 
 is a simple teletype, whereas the SUPDUP defines a display terminal is a simple teletype, whereas the SUPDUP defines a display terminal
Line 190: Line 194:
  
 The trade-off associated with SUPDUP is that the remote operating system must be able to talk to SUPDUP's virtual terminal. Alas, most operating systems understand just printing terminals, leaving all other tasks to user programs. Since the SUPDUP's virtual terminal is not a superset of the ordinary teletype, such operating systems cannot communicate with it Instead, user programs must create a SUPDUP server. To put this in perspective, the SUPDUP protocol is meant as The trade-off associated with SUPDUP is that the remote operating system must be able to talk to SUPDUP's virtual terminal. Alas, most operating systems understand just printing terminals, leaving all other tasks to user programs. Since the SUPDUP's virtual terminal is not a superset of the ordinary teletype, such operating systems cannot communicate with it Instead, user programs must create a SUPDUP server. To put this in perspective, the SUPDUP protocol is meant as
-a meet for having to know exactly everything about all types of terminals connected to a system; the annoying symptom of this is the PRIMOS Emacs' prompt for supplying the terminal type the user is on, or the Termcap feature of Unix.+a ment for having to know exactly everything about all types of terminals connected to a system; the annoying symptom of this is the PRIMOS Emacs' prompt for supplying the terminal type the user is on, or the Termcap feature of Unix.
  
  From the user's point of view, SUPDUP is wonderful. It "magically" provides the right translations so that a remote login user need not manually set switches every time a remote login to a different host  From the user's point of view, SUPDUP is wonderful. It "magically" provides the right translations so that a remote login user need not manually set switches every time a remote login to a different host
Line 198: Line 202:
  
  SUPDUP also provides flow control for non-networked terminals in a manner superior to that of the XON/XOFF mechanism which robs the user of all editing commands using CONTROL-S and control-Q. It is akin to the flow control mechanism of TCP. It works by means of allocation, which is stored in the host, and which is the number of  SUPDUP also provides flow control for non-networked terminals in a manner superior to that of the XON/XOFF mechanism which robs the user of all editing commands using CONTROL-S and control-Q. It is akin to the flow control mechanism of TCP. It works by means of allocation, which is stored in the host, and which is the number of
-characters~ thatcan be accommodated by the terminal short of overloading it, thus it acts much like network data layer windowing.+characters that can be accommodated by the terminal short of overloading it, thus it acts much like network data layer windowing.
  
  The graphics protocol of SUPDUP has the nice property that the terminals using it are not aware of whether they are used remotely or locally. The graphics connection does not need any additional network connections, nor is there any need for preparation--graphics may be initiated any time during text transmission. The graphics protocol can work with either display list or bit-matrix terminals, allowing, for example, drawing pictures on part of the screen and keeping the rest of it for text display without mixing the two.  The graphics protocol of SUPDUP has the nice property that the terminals using it are not aware of whether they are used remotely or locally. The graphics connection does not need any additional network connections, nor is there any need for preparation--graphics may be initiated any time during text transmission. The graphics protocol can work with either display list or bit-matrix terminals, allowing, for example, drawing pictures on part of the screen and keeping the rest of it for text display without mixing the two.
Line 216: Line 220:
 not prompt for password again. While this may not be a good feature in a security-minded environment, it certainly makes life easier and faster when one has to repeatedly remote login through layers of remote hosts (a practice normally discouraged for efficiency reasons), or when repeatedly opening sessions to the same remote host. not prompt for password again. While this may not be a good feature in a security-minded environment, it certainly makes life easier and faster when one has to repeatedly remote login through layers of remote hosts (a practice normally discouraged for efficiency reasons), or when repeatedly opening sessions to the same remote host.
  
-===== User's impressions of the Chaosnet/SUPDUP MIT environment =====+===== - User's impressions of the Chaosnet/SUPDUP MIT environment =====
  
  There was a time when MIT AI Lab's computing environment was regarded as the best in the world. That assessment is no longer uttered by quite as many people these days, and I think the main problem is the fact that the computing environment is outgrowing its local area network.  There was a time when MIT AI Lab's computing environment was regarded as the best in the world. That assessment is no longer uttered by quite as many people these days, and I think the main problem is the fact that the computing environment is outgrowing its local area network.
Line 227: Line 231:
 a virtual file system is becoming more painful each year, especially to new users. a virtual file system is becoming more painful each year, especially to new users.
  
-  The robustness of Chaosnet, which is the flip-coin side of the simplicity of its design, is unquestionable. In my one year at the AI Lab I do not recall a single instance of the net being down. (Though, I do remember various bridges giving out from time to time, but no more or less often than "normal" PDP-11's.)+The robustness of Chaosnet, which is the flip-coin side of the simplicity of its design, is unquestionable. In my one year at the AI Lab I do not recall a single instance of the net being down. (Though, I do remember various bridges giving out from time to time, but no more or less often than "normal" PDP-11's.)
  
- One also misses the ingenious behavior of SUPDUP once one grows accustomed to the conveniences it provides. Using things called "Emacs" under Unix (or VMS) and Sytek has a positively neolithic feel to it in comparison to Emacsing on Chaosnet under SUPDUP.+One also misses the ingenious behavior of SUPDUP once one grows accustomed to the conveniences it provides. Using things called "Emacs" under Unix (or VMS) and Sytek has a positively neolithic feel to it in comparison to Emacsing on Chaosnet under SUPDUP.
  
 The remote login facility works very nicely, as does the file transfer protocol, provided the two do not go head-to-head in The remote login facility works very nicely, as does the file transfer protocol, provided the two do not go head-to-head in
Line 235: Line 239:
  
  The Chaosnet's main tenet, that of providing the lisp machine user with an umbilical cord to its mother file server, is proving itself every day. It is only when the network is stretched past its original design goal to accommodate terminal traffic that the user has any reason to complain.  The Chaosnet's main tenet, that of providing the lisp machine user with an umbilical cord to its mother file server, is proving itself every day. It is only when the network is stretched past its original design goal to accommodate terminal traffic that the user has any reason to complain.
 +
 I am convinced that the bulk of what local networking does at I am convinced that the bulk of what local networking does at
 MIT is done if not optimally then at least satisfactorily. Considering the huge number of interfaces to so many different operating systems and peripherals, it is a winning system. However, considerable changes will need to be made in the way interactive computing is handled MIT is done if not optimally then at least satisfactorily. Considering the huge number of interfaces to so many different operating systems and peripherals, it is a winning system. However, considerable changes will need to be made in the way interactive computing is handled
Line 240: Line 245:
 deteriorating. Who knows, it might be even possible to improve it. deteriorating. Who knows, it might be even possible to improve it.
  
-===== Other network projects at MIT =====+===== - Other network projects at MIT =====
  
  Currently, MIT at large is developing a personal workstation net consisting of IBM PCs, DEC VAXes, DEC Professional microcomputers, and the Ethernet network, which will provide service to the thousands of faculty and students of the Institute. This project is called Project Athena, and it will be gatewayed to the Chaosnet.  Currently, MIT at large is developing a personal workstation net consisting of IBM PCs, DEC VAXes, DEC Professional microcomputers, and the Ethernet network, which will provide service to the thousands of faculty and students of the Institute. This project is called Project Athena, and it will be gatewayed to the Chaosnet.
Line 252: Line 257:
  Independently of the Project Athena, both of the MIT computer labs are putting in new, separate Ethernets meant to install in every LCS office a bit-mapped display to be used with a time-sharing CPU such as  Independently of the Project Athena, both of the MIT computer labs are putting in new, separate Ethernets meant to install in every LCS office a bit-mapped display to be used with a time-sharing CPU such as
 the MIT-XX, a DecSystem-20, or a Vax 750 under 4.2bsd Unix in single user mode, and likely a TI Explorer Lisp Machine, and at the AI Lab, a 3640. the MIT-XX, a DecSystem-20, or a Vax 750 under 4.2bsd Unix in single user mode, and likely a TI Explorer Lisp Machine, and at the AI Lab, a 3640.
- +I think this is the right way to go. This would enable putting the networked terminals only on a subnet, possibly improving response time for all, distributing the known bottlenecks over more cable. The trend
-  think this is the right way to go. This would enable putting the networked terminals only on a subnet, possibly improving response time for all, distributing the known bottlenecks over more cable. The trend+
 to have people work all day on dedicated lisp machines of a Symbolics 3640 class brings us one more step out of the dark ages of interactive computing, the age of the hardwired time-shared terminal. And, with projects such as the Gerald S. Sussman's current attempted construction of a lisp machine suffused with RAM (eliminating all need for virtual memory management), to have people work all day on dedicated lisp machines of a Symbolics 3640 class brings us one more step out of the dark ages of interactive computing, the age of the hardwired time-shared terminal. And, with projects such as the Gerald S. Sussman's current attempted construction of a lisp machine suffused with RAM (eliminating all need for virtual memory management),
 and with the already manifested trend to have users save files on their local lisp machine's secondary storage (disk or whatnot), we are compelled to change our view of what the mission of networks is from that of helping make do with resource limitations to that of being mere passageways. and with the already manifested trend to have users save files on their local lisp machine's secondary storage (disk or whatnot), we are compelled to change our view of what the mission of networks is from that of helping make do with resource limitations to that of being mere passageways.
  
-===== Bibliography =====+===== - Bibliography ===== 
 In order of utilization by this paper: In order of utilization by this paper:
  
Line 266: Line 271:
  
  "A History of Networking at MIT LCS/AI Lab", an LCS seminar by Dave Clark, MIT Lab for CS staff member, Massachusetts Institute of Technology, summer 1984.  "A History of Networking at MIT LCS/AI Lab", an LCS seminar by Dave Clark, MIT Lab for CS staff member, Massachusetts Institute of Technology, summer 1984.
 +
 "The Lisp Machine Manual", David Weinreb and David Moon, internal  "The Lisp Machine Manual", David Weinreb and David Moon, internal 
 publication, Massachusetts Institute of Technology, The AI Laboratory. publication, Massachusetts Institute of Technology, The AI Laboratory.
Line 272: Line 278:
  
  "C690 Seminar: Computer Networks and Communications", Professor Frank F. Prosser, Fall Semester1984, Indiana University Computer Science Dept.  "C690 Seminar: Computer Networks and Communications", Professor Frank F. Prosser, Fall Semester1984, Indiana University Computer Science Dept.
-Personal Communication sources, in alphabetical order: Alan Bawden, MIT AI Lab, September 1983 - August 1984. + 
-Bernie Greenberg, Symbolics Inc., September 1983 - August 1984. Scott A. Jones, MIT AI Lab, August 1984 - May 1985. +Personal Communication sources, in alphabetical order: 
-Christopher Lindblatt, ibid, August 1984 - May 1985. David Plaisier, Indiana University CS Dept., June 1985 Jerry Roylance, MIT AI Lab, September 1983 - August 1984. Steve Strassmann, ibid, September 1983 - December 1984. Jonathan Taft, ibid, September 1983 - August 1984. +  * Alan Bawden, MIT AI Lab, September 1983 - August 1984. 
-Gail Zacharias, lipid, December 1984 - January 1984.+  * Bernie Greenberg, Symbolics Inc., September 1983 - August 1984.  
 +  * Scott A. Jones, MIT AI Lab, August 1984 - May 1985. 
 +  * Christopher Lindblatt, ibid, August 1984 - May 1985.  
 +  * David Plaisier, Indiana University CS Dept., June 1985  
 +  * Jerry Roylance, MIT AI Lab, September 1983 - August 1984.  
 +  * Steve Strassmann, ibid, September 1983 - December 1984.  
 +  * Jonathan Taft, ibid, September 1983 - August 1984. 
 +  * Gail Zacharias, lipid, December 1984 - January 1984.
  
  The author wishes to thank Professor Marvin Minsky for the invitation-to and generous support while at the MIT AI Lab, September 1983 to August 1984, and to Professor Douglas R. Hofstadter for making it possible. Th research conducted at The Massachusetts Institute of Technology's Artificial Intelligence Laboratory is supported by The Office of Naval Research, United States Department of Defense. The views expressed herein are those of the author alone and the references indicated above do not imply anyone's endorsement of these views. The  The author wishes to thank Professor Marvin Minsky for the invitation-to and generous support while at the MIT AI Lab, September 1983 to August 1984, and to Professor Douglas R. Hofstadter for making it possible. Th research conducted at The Massachusetts Institute of Technology's Artificial Intelligence Laboratory is supported by The Office of Naval Research, United States Department of Defense. The views expressed herein are those of the author alone and the references indicated above do not imply anyone's endorsement of these views. The
tr173.1608137268.txt.gz · Last modified: by victor
Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki