MIT-MC.ARPA – INTERIM COMPUTER MUSEUM

MIT-MC.ARPA was a DECSYSTEM-10 KL10 Model 1080 operated by the MIT MACSYMA Consortium, serving as one of the most significant systems in the development of AI and symbolic computation during the 1970s and early 1980s. Running the Incompatible Timesharing System (ITS), MC provided a uniquely open and collaborative computing environment where many pioneering software systems were developed. As a major ARPANET node, it allowed remote access and encouraged experimentation by a community of programmers, researchers, and students. The system’s flexibility and accessibility made it a central platform for advancing interactive computing and large-scale symbolic programming.

One of the most influential software systems developed and run on MIT-MC was MACSYMA, a powerful symbolic algebra system. It was originally developed from 1968 to 1982 at MIT‘s Project MAC. David Moon played a critical role in MACSYMA’s implementation, contributing both to its underlying architecture and the development of MACLISP, the Lisp dialect it relied on. Moon’s work on garbage collection, performance tuning, and the system’s internal representation of algebraic structures helped make MACSYMA one of the most advanced symbolic manipulators of its time.

The project was initiated in July, 1968 by Car Engelman, William Martin, and Joel Moses. Engelman and Martin were largely responsible for the front end, expression display, and polynomial arithmetic. Moses implemented the simplifier, the indefinite integration support, and Macsyma heuristics, including Risch support. Martin was in charge of the project until 1971, and Moses ran it for the next decade. The integration of sophisticated mathematical capabilities with efficient system design demonstrated the potential of symbolic AI and laid groundwork for future mathematical software. MACSYMA became a landmark achievement in symbolic computing.

Macsyma was initiated by MIT’s Project MAC, established in 1963 within the Department of Electrical Engineering (later renamed to the Department of Electrical Engineering and Computer Science). Funding was provided by the Defense Advanced Research Projects Agency (DARPA). Project MAC became became MIT’s Lab for Computer Science (LCS) in 1976. One of Project MAC’s major undertakings was the development of Macsyma. Development began on a PDP-10 KA, running ITS (Incompatible Timesharing System). Eventually, due to increased memory needs, Project MAC purchased from Digital Equipment Corporation (DEC) a PDP-10 KL around 1971. This machine was named MIT-MC, where MC stood for Macsyma Consortium. It also ran the ITS operating system, as did the other PDP-10s at MIT’s AI Lab, Math Lab, and Dynamic Modeling Group. MIT-MC’s KL10 had serial number 1038, and was in service at MIT from 1971 to 1984.

When the Project MAC upgraded the Macsyma development machine from a KA10 to a KL10 (which had a different microarchitecture and instruction timing), they discovered that MACSYMA’s performance suffered significantly. This was primarily because the KL10, being optimized for throughput and virtual memory, handled certain low-level operations (like stack manipulation and context switching) less efficiently than the KA10 in the specific ways MACSYMA relied on.

To address the Macsyma performance issues that resulted from the change from a KA10 to a KL10, MIT engineers—particularly David Moon and Tom Knight—designed a hardware workaround they called the KL-udge. It was a small piece of additional logic added to the KL10 that improved performance for MACSYMA and certain other Lisp programs by modifying how stack frames and interrupt handling worked, effectively emulating KA10-like behavior where needed. The KL-udge allowed the AI Lab to continue running their complex symbolic computations without rewriting MACSYMA for the new architecture, preserving both performance and compatibility.

KL-udge (a play on the word kludge, meaning a workaround or hack) was a nickname given by MIT hackers to a custom hardware modification created to support the MIT Lab for Computer Science’s MACSYMA system on the KL10 processor of MIT-MC.ARPA.

The KL-udge is a prime example of the MIT hacker ethic: elegant or not, the goal was to make the system work better, even if it meant bending the rules of conventional hardware design.

MIT-MC was added to the ARPANET in 1976. The email message from Dave Moon, below, announces this event:

Date: 14 FEB 1976 0035-EST
From: MOON at MIT-MC
To: NETWORK-LIAISON-GROUP at MIT-MC, *** at MIT-MC
CC: JM at MIT-MC, Feinler at BBN-TENEXB, JM at MIT-ML

MIT-MC (the Macsyma Consortium KL10) is now on the Arpanet
as host number 354 (octal) or 236 decimal. Note that this
is DIFFERENT FROM the host number in the Resource Notebook.

The following excerpt from a April 1980 Project MAC report, written by Ellen Golden, includes:

The primary hardware improvement this past year was the addition of a million words of memory to the MC machine, bringing the total available memory
to 1.5 million words. The primary effect of this addition was
to increase the ratio of cpu time to connect time by about 50%.
A secondary effect was that the number of connect hours increased, probably due
to the enhanced performance of the system.

In order to handle the increased demand for secondary storage, a new 300 megabyte disk drive will soon be added to the 3 RP04s now online.
In addition, a LISP Machine (developed by the M.I.T. Artificial Intelligence
Laboratory) will be made available for use by MACSYMA users.