An Early Local Area Network

I graduated from Carnegie-Mellon University in the spring of 1973, and through an old roommate, I managed to get a summer job at Bell Labs, Department 1227 (Acoustics Research), working for Cecil Coker. My job was to write a command-language interpreter for the DDP-516 computers the department used as workstations, so that users didn't have to use the switch register and bootstrap to launch each application they wanted to run.

The workstations in the Acoustics Research Department were connected by an early local area network based on J. R. Pierce's slotted ring network architecture. This is described in a series of 3 articles in the Bell System Technical Journal, 50^th anniversary edition (July-August, 1972):

• Network for Block Switching of Data, by J. R. Pierce, pages 1133-1145.
• An Experimental Data Block Switching System by W. J. Kropfl, pages 1147-1165.
• An Experimental Interconnection of Computers Through a Loop by C. H. Coker, pages 1167-1175.

In short, the J. R. Pierce's article describes a proposal for a national-scale data network consisting of a hierarchy of slotted loop networks, with the option to add redundancy. Pierce is somewhat famous for having, among other things, coined the word transistor and for overseeing the development of the first commercial communications satellite Telstar I.

Kropfl's paper describes the design and construction of a a prototype demonstrating Pierce's loop network using T1 lines with a bit rate of 1.544 MHz.

Cecil Coker's article builds on Kropfl's paper to describe using Kropfl's network interface boxes to connect a pair of DDP-516 computers. Coker demonstrated the utility of the loop network with an application to copy data from the local disk of one computer to the other over the network.

By the time I got to Bell Labs, the newtork had grown from 2 identical machines to 3 identical DDP-516 workstations and a final DDP-516 running as a disk server. The machines occupied two rooms across the hall from each other and down a long hall from the Bell Labs computer center. While I was there, a CSP-30 compute server was added, along with a TI-980 minicomputer. As I left Bell Labs, I was given a copy of the BSTJ 50^th anniversary edition, and I wrote down the following diagram of the network:

My notes made in the summer of 1974 showing the network as it was then.
 

Looking back at this system from half a century later, it is quite an amazing mix. It was among the very first local area networks connecting worstations to a disk server and compute server. The idea of a workstation with mouse and graphics display was still quite novel. The workstations were, however, tiny, with only 32K bytes of main memory. We still used punched cards and rarely paper tape for programming, and nowhere did I see any hint of a file system. Instead, people used disk track numbers for programs and data.

The three DDP-516 workstations were not physically small. The computer itself was the size of a desk, countertop high, with a white formica top and a bright orange body (Honeywell's trademark color). The fans were loud enough that the labs built a soundproof wall between the computer room and the work area. Room 2D-518 (on the 5^th floor of building 2D) held the disk server computer, 4 top-loading disk drives, and the computer for one workstation, the one I always used. The back of the room, behind a soundproof wall, held all the I/O devices for the workstation. I found this Bell Labs publicity photo of that workstation (from before they built the soundproof wall):

WORD MERCHANTS — Bell Laboratories scientists R. W. Schafer, J. L. Flanagan, and L. R. Rabiner (left to right) listen to synthetic speech produced by their new technique that uses only one fiftieth the amount of digital information previously needed for computer generated speech. Schafer (foreground) controls the graphic representations of synthetic speech patterns displayed on the screen to his left. Synthetic speech, produced in this manner, may one day provide information to people who dial telephone numbers that have been disconnected or changed, or other services such as up-to-the-inute weather reports and telephone numbers.
 

James L. Flanagan was the head of the Acoustics Research Department, and I remember that Larry Rabiner was deeply involved in both speaker recognition and speech recognition — two distinct problems, since you can recognize who is speaking without knowing what they say, and you can recognize words without knowing who is saying them.

In the photo, Schafer is using a mouse. It was a home-made mouse, an aluminum block holding two shaft encoders at right angles, with a push button on the far end (out of sight). There was no mouse ball. Instead, the two perpendicular shaft encoders had polished steel disks that rolled or skidded on the lap board as you moved the mouse.

Rabiner is resting his elbow on the paper-tape punch of a model 33 ASR Teletype, the primary text I/O device of the workstation. The three men are wearing headphones so they can listen to the audio output without being distracted by machine noises. The audio amplifier driven by the computer's digital to analog convereter and a loud speaker are on the shelf above the display scope.

The scope sits next to the front panel lights and switches for the DDP-516. The actual computer is behind Rabiner, off the image to the right. There's a high-speed paper-tape reader partly obscured by Flanagan, and a Documation punched-card reader.

During the summer of 1973, I used the Teletype in the photo, but they replaced it with a Diablo HiType daisy wheel terminal by 1974. That was both quieter and faster (300 baud instead of 110 baud).

When I came there in 1973, you would set, in binary, the number of the disk track you wanted to run on the switch register, then hit master clear and run. The bootstrap program, in write protected locations 1 to 20₈, would load the linking loader from disk, then launch the loader to read from the indicated track number. All the software was originally paper-tape based, but the copies on disk had been patched to read and write from disk. There was no file system.

To run a FORTRAN program from punched cards, you'd load the program in the card reader, enter the number of the compiler on the switch register, hit master clear and run. The compiler was patched to write object code to a dedicated disk track, and the listing to another dedicated track. Then you'd enter the number of the linking loader and launch it. The loader always searched the library on disk, and it was a load-and-go loader, so your program would run.

All 4 DDP-516 systems had just 16K 16-bit words of 0.96μs core memory. These were single accumulator single-address machines where the top 2 bits of each instruction indicated the addressing mode (an indirect bit and an index bit). This seems incredibly small and slow by modern standards, but it was enough to do some fairly interesting work. The machines had DMA channels, necessary for the high speed of the Pierce loop network as well as for effective use of the disk drives.

Each DDP-516 had a fixed-head disk. The computer could, of course, access any track, but two tracks were also connected to the scope interface and used as backing storage to refresh the scope display. The scope interface had one bit to control which track was currently being displayed, and you could do animation by writing one track while displaying the other. I am working from stale memory as I write this, but I think the scope display worked by starting at address 0,0 at the start of each disk revolution. The data on the display track was interpreted as a sequence of display commands with just a few bits (perhaps 4) per command: One bit meaning beam on or off, one bit meaning increment and decrement X, one meaning increment or decrement Y, and one bit meaning step size (small or large).

The fixed-head disks weren't big, just one megabyte (or half a million 16-bit words). That is where the loop network became important. Before the loop network, backups were done to 9-track tape, but with the loop network, one DDP-516 became a disk server, with 4 top-loading disk drives, each the size of a washing machine, each holding a removable 10-surface CDC disk pack. A network application ran permanently on the disk server machine waiting for requests to read or write a disk pack. An application on the workstations allowed any track of the local disk to be copied to or from the disk pack. This application was almost certainly only a small extension of the one described in Coker's BSTJ article.

The acoustics research department was interesting for other things. I was at Bell Labs during the Watergate hearings in the summer of 1974. I lived with my former roommate from Carnegie-Mellon University, and we'd go home every evening and eat dinner, transfixed by the news coverage of the hearings. James Flanagan, the department head, was one of the people who'd analyzed the 18-minute gap in the Nixon Tapes, and it was his analysis that showed that the gap was produced by multiple erase and rewind operations. Clearly, whatever President Nixon had said during those 18 minutes was something that needed to be carefully and thoroughly erased.

Larry Rabiner was very interested in the speaker recognition problem, and he had just about everyone in the lab record the sentence "we were away a while ago" for testing. That sentence has almost no stops, it's almost entirely open-voiced, and speaker identity is most clearly conveyed by open voice speaking.

My boss, Cecil Coker, was really interested in physical acoustics of the vocal tract. He had speech output systems based on simulation of the physical vocal tract, with inputs controlling tongue, pharynx and jaw positions, and he wanted to validate his models by getting volunteers to allow microphones to be stuck down their throats (or into their throats through the skin) while they spoke. His software used the graphics display on the DDP-516 workstation to show a cross-section of the vocal tract that you could manipulate with the mouse while you listened to the sound it produced.

Coker and many others in the department were also very interested in how to efficiently compress speech and other audio data for digital storage and trasmission. This obviously relates to the business of the telephone company, and after I left Bell Labs, James D. Johnston, another former roommate of mine went there and took up the compression problem with a vengance, helping-invent the encoding techniques that evolved into the ubiquitous MP3 data format.

Of course, there was other interesting stuff going on at Bell Labs. The vending machines were one floor up from my lab, and when I went up for an apple (my most common midafternoon snack at the time), I'd peer into the room where Unix was being developed.

I met Ken Thompson on several of my trips to the computer center. I used the big Honeywell 9000 mainframe as a text editor to prepare code to run on the DDP-516. Better to edit text on line than to keypunch everything. Ken was frequently in the terminal room working on his chess program, a distant ancestor of Belle.

I also met Max Mathews, he was the director of the Acoustical and Behavioral Research Center at Bell Labs, that included Flanagan's department, and he's famous as a pioneer in digital music.

Outside the labs, my roommate and I went into New York one weekend and spent several hours with Arthur C. Clarke at his hotel. It turns out that being two young engineers from Bell Labs put us high enough on the pecking order that he was happy to see us. My roommate had a nice new HP35 pocket calculator, then still new enough that Clarke had never seen one before. Clarke had a wonderful bit of advice, that I'll paraphrase here:

In any work of fiction, the author is entitled to one outrageous assumption, and it doesn't matter how outrageous. So long as everything in the story follows from that assumption and the way the world really works, the audience will go along with you. Make any additional assumptions, though, and you'll lose your audience."