[ih] XEROX/PUP and Commercialization (was Re: FYI - Gordon Crovitz/WSJ on "Who Really Invented the Internet?")

[Marty Lyons]

On Jul 29, 2012, at 8:14 PM, Dave Crocker wrote:

> I don't believe for one minute that X.25/X.75 were capable of the kind of usage TCP/IP has experienced.  The complexity and limitations they imposed seem to me to require massively more expensive and massively less flexible, robust, etc., etc., operation.

In the U.S. some of the X.25 universe of companies included Telenet (last owner: Sprint), Uninet (part of Sprint circa 1986), and Tymnet (last owner: MCI/Verizon).  There were some other networks as well, but Telenet and Tymnet continued operating the longest until being replaced by the IP backbones of their parent companies.  I'm not sure of the final dates but I seem to recall sometime around 1997 the last X.25 backbones were withdrawn for new connections.  

One of the principle reasons X.25 failed was the utter complexity of getting reliable and fairly priced billing contracts.  Back in 1982 the lab I worked at ran both a Telenet and Uninet node, and the bills were very expensive and difficult to verify for correctness.  Around 1985, I got the same lab connected to BITNET on a 9600bps leased line, and the accounting was simple -- one flat rate per month.  BITNET took off in terms of universities joining because all you needed was two CSUs, a friendly partner site to connect to, and a little bit of money for a circuit (this was still in the era of requiring government sponsorship for ARPA connectivity, so this was the easiest way to get a dedicated campus connection). Almost every BITNET site ended up with a fully loaded circuit very quickly, particularly when the undergrads discovered online chatting (see stories elsewhere of how BITNET RELAY almost shut down the network due to load).

In 1993, building the network at America Online I found that our X.25 providers were sending us massive bills and we still couldn't get a handle on the finances (ten years after my university days and apparently nothing had changed).  We moved as fast as possible to trunk all the traffic on IP backhaul.

So one argument in why the Internet (in all its component parts) has succeeded was the simplicity of bit accounting -- get a leased line, and pay one price for the connection per month, independent of loading of the circuit.  Buying channels or a full T-span has always made much more sense than paying a X.25 carrier to account for your packets, route them correctly, and manage load balancing in their cloud -- which never seemed to be handled well, since the routing tables had to be updated by hand and pushed to the control nodes.

TCP/IP moved us from a world of "the phone company" where the model was charge-by-packet and let someone else do network control, to simple finances and control of our own network management. There was just no way financially or administratively X.25 stood a chance once NSF dropped the AUP, and people could buy a Cisco router, call UUnet/CIX member and get connected directly.  X.25 was the mainframe of the networking world (centralized control), TCP/IP was the IBM PC (decentralized).

Around 1995, I had conversations with several (different) IP backbone providers which were trying to figure out how to sell connections with bit rate billing. Thankfully that never happened. I got the impression that some were close to floating it as a product though.

Having run networks of every one of those "other" protocols (SNA, DECnet, XNS, SPX/IPX), it was a happier time when they all finally seemed to go away.  Of all of the alternatives, XNS was probably the most interesting and if Xerox had licensed it correctly, it may have had wider use.  And SNA continues to live on in private environments -- some quite substantial.

Side story on deployment in interesting places: I converted all of South Pole Station, Antarctica to TCP/IP in November 1991.  Previously the station used a mix of various protocols.  The Suns were ready out of box, but once the Vaxen has TGV MultiNet installed, the pole was free from gulag DECnet and we enabled lots more science to happen by installing freeware IP clients on PCs and Macs.

Marty