A Correspondence on LAN History

On Wed, 07 Mar 2007 15:47:45 -0500 Craig Partridge wrote:

I see from your signature that you state that 2007 is the 30th year of local area networking.

I'm curious why you chose 1977 as the starting date. The first Ethernets were earlier (the key paper was published in July 1976 in CACM and Bob Metcalfe apparently had the scheme largely complete by 1974).

Ethernet was essentially little more than a wired implementation of the Aloha system (interestingly, originally intended as a wireless network... indeed, things have come full circle) developed at the University of Hawaii.

Bob Metcalfe's original Ethernet implementation was a hopelessly impractical system... operating at 2 megabits, using clumsy thick-wire linear-bus coax and "vampire" taps, plagued by reflections which required taps be placed at wavelength multiples, and leaving nasty physical holes in the side of the cable if a tap were ever removed. The entire network was disrupted when taps were added or removed, and identification and isolation of faults was a nightmare which got worse as the bus got longer. Moreover, the design called for "blind" transmission of data blocks, where the sender had no idea if the intended destination were even present, let alone with a receiver enabled and incoming buffer ready to receive it. After transmission was accomplished, the sender had no idea if the transmission had been received, until (hopefully) the expected response was received or (eventually) a timeout of some kind would occur. If no response was returned, one didn't really know if it was the transmission, or the reply, which had been lost (and this greatly complicated attempts at error recovery). The original spec simply said (hopefully) that such problems would be "handled by the higher level protocols".

The DCS ring (done by Dave Farber's team at Irvine) was published in 1975 and I assume working well before.

There were numerous LAN cabling systems and protocols being used on an experimental basis at universities and research centers... I believe that the Aloha system actually was first reported sometime in the late 1960's. And of course, other (basically point-to-point) communications (both asynchronous and synchronous) had been in use for quite some time in a variety of implementations.

So I'm curious -- what even caused you to pick 1977?

The first commercially available local area network was Datapoint's "The ARC System", based on a hardware cabling system which was called "ARCnet". The system incorporated a number of key advances:

1. a true packet-based protocol using variable-length packets, with the hardware managing transmission and most of the error checking, including packet buffering and the host-free self-configuring token passing scheme used to coordinate network access;
2. a simple, robust cabling scheme based on already widely used RG-62U coaxial cable and BNC connectors (the same cabling system used by both IBM 2260 and 3270 terminals, the most widely used business terminals at the time)... cabling readily available worldwide at low cost in a large variety of styles (direct burial, overhead, plenum, etc etc);
3. a far more practical "interconnected stars" cabling topology based on "active hubs" (Ethernet was never really practical until Ethernet finally adopted ARCnet-style "interconnected stars" cabling topology, which they didn't do until after a ultimately short-lived attempt to use nearly-as-flawed "cheapernet" thin-coax cable, but still using the fundamentally flawed linear bus scheme);
4. ARCnet ran at 2.5 megabits, but its more efficient protocols and smaller packet overhead yielded nearly twice the per-megabit performance in practice than Ethernet did;
5. The ARC System when first delivered (to Chase Manhattan Bank, towards the end of September 1977) was not just a multi-megabit packet bus... it was a full-fledged commercially ready product, fully supported by disk operating systems the bank was already using and the high level business programming language the bank was already using for their applications (Ethernet wasn't available as a commercial product, and even then with limited OS and applications programming language support, until several years later);
6. The system as shipped in September 1977 already included an innovative, fully-distributed, multilevel enqueue system supporting deadlock-free multi-file locking across multiple volumes, servers, and even conjoint cabling systems... which was supported both at the operating system and high-level-programming-language levels;
7. The ARC System also provided for an innovative new approach (still very widely used in nearly all LAN software to date) to conceptualizing access to remote resources by mapping entire remote disk volumes into imaginary "local disk drives", allowing programs to transparently access files on remote volumes as easily as they could reach out and access (or even create) files on an assortment of purely local disk drives;
8. Mapping of these remote volumes, and establishing access permissions based on user names and passwords, was thus accomplished OUTSIDE the applications (and thus transparently to applications) and allowed many existing programs to work without modification in the LAN environment. (This compares with "DECnet-style" networking, where programs had to all be rewritten to use special remote-file opens specifying server names, user names, passwords, and other information not widely written into existing applications at the time);
9. You didn't even need to know on which server your volumes were presently mounted; you simply specified the name of your volume and your user name and password; the system would locate (on any server on any cabling system you were connected to) that named volume which your user name was allowed to use. (Even most LAN schemes today still don't provide that feature). Disk volumes could be moved (while not being used, of course) from one server to another, transparently to applications;
10. The ARC System, announced on December 1, 1977 (interestingly, as little as one month before announcement Datapoint had been planning to call their new product "INTERNET" (!)) was not just a research project in a lab, but had already been operating in production in a commercial (banking) customer environment for more than two months, and could be ordered as a shipping, production product (and fully ready for applications programs to use it immediately) for essentially immediate delivery.
11. Moreover, the original shipping version of The ARC System used the exact same DOS.D 2.2 operating system that most larger Datapoint customers were already using on their production systems. This was not a "do it yourself kit" requiring applications programmers to "reinvent" wheels to adapt their programs to a completely alien new environment. This LAN system worked with existing multiuser business applications and transaction processing on (with the exception of the new ARCnet cabling system itself) existing hardware.

I'll also mention, for the record, that preproduction ARCnet hardware (which at the time were called FRIL comboxes, where FRIL stood for Fast Resource Intercommunication Link) were operational in early 1976, but Datapoint wasn't really sure what to use them for. Initial inhouse programs we were using with FRILs centered on Telnet-style remote program execution, FTP-style file transfers, or printer sharing... the same sorts of things that early ARPAnet was primarily being used for. I proposed what ultimately developed into The ARC System in summer 1976. The project was quickly approved, and I started writing the client-end by September that year. By the end of the year, the first-generation client-end software was ready for testing. I started coding the Server immediately thereafter and I was using the first versions of the client and server software for my own inhouse elaboration of the system by February or March 1977. It was quickly adopted through Advanced Product Development and then into Software Development quickly afterwards. Our Marketing Department got their installation in mid-Summer and they arranged for Chase Manhattan, a key Datapoint customer, to get the first out-of-house installation, which both I and David Hovel (who did the changes to our Datashare multiuser business programming language) personally went to New York to install in late September 1977. The public announcement of The ARC System took place in New York City, in combination with Datapoint's annual stockholder's meeting, on December 1, 1977. The announcement took up most of the front page of the immediately subsequent issue of Computerworld. :-)

Thanks!

Craig Partridge

You're welcome, Craig! And thanks for asking. Nice to get the opportunity to help set the historical record straight.

Use the BACK control on your Web browser, or click here to return to my main home page.

This page and all linked contents originating with me are Copyright (C) 1995-2007 by Gordon E. Peterson II, all rights reserved worldwide. This page last revised March 8, 2007.