[ih] Why did TCP win?

[Bill Nowicki]

Here are some of my anecdotes. Forgive me if you have already heard these stories, since we rehash this every year or so.
Although InterOp was the show that made money and got famous, it was mostly for routers and switches. Although it is my bias, all those switches would not have been useful without the end-user applications and devices that plugged into them. At Sun we sponsored an event called "Connectathon" that was by engineers for engineers, so got little marketing attention. But did mean that the entire stack (at least for file service) was tested periodically, and we found new issues each year. Alas, Sun marketing was convinced that customers really wanted that STREAMs interface so they could use all those virtual-circuit protocols. There was a large group funded to develop the OSI protocols, and an even larger one to emulate the Socket interface. Meanwhile they tolerated me trying to help the research community a bit.
One story is about Cisco being "multi protocol" (Len Bosack was a good friend of mine in grad school). Most customers had TCP/IP plus one other one. If you were a DEC shop (and not Unix) then it was DECNet, etc. While the Internet evolved to be how those islands inter-connected. They lasted quite a while, however. I remember just after I got frustrated and left Sun, I got a call from Sun marketing. They were looking for customers selling applications to use the "protocol independent" RPC package instead of the Socket-based one we used at Legato (so we could use all those other protocols too). I could not convince him that we only wanted one protocol at a time; it was hard enough testing all those combinations, and the testing and troubleshooting of subtle bugs was the hardest part. This was in 1989.
I remember a talk by Greg Chesson in about 1983. Greg said that Ethernet and TCP/IP would never catch on because its addresses were in binary. DataKit addresses were in decimal, which he said made it so much easier to use. I could never tell if Greg was serious or trolling. AT&T was still pushing for those phone-like protocols in 1989. One of AT&T's "requirements" was that a new implementation of TCP/IP must be able to be dynamically loaded into the Unix kernel without interrupting any connections in progress, since that was how their phone switching software worked. That requirement along with many others was never met. My guess was that other countries had similar ideas in their phone monopolies. The irony, of course, is that Greg had the last laugh. If I want to send someone a quick message now days, I do not ask for their X.500 distinguished name, but their phone number.
Bill Nowicki    On Monday, February 3, 2025 at 01:19:38 PM PST, Olivier MJ Crépin-Leblond via Internet-history <internet-history at elists.isoc.org> wrote:  
 
 

On 03/02/2025 20:09, Noel Chiappa via Internet-history wrote:
> The period when TCP/IP was used inside organizations which had no extrnal
> connection (and thus could have used_any_ protocol family) is thus probably
> instructive. That period was short, but did exist - Novell Netware was a big
> hit during it. (At a point in time when PC's and LANs were becoming popular.)
>
> One thing that would have made TCP/IP popular at that point was that it was
> available for so any kinds of machines (for reasons touched on earlier); it
> wasn't 'company X's protocol'.

Bravo. that's the closest to my experience in the UK.
In the mid-nineties in Imperial College we asked ourselves why the 
sudden change of direction from OSI to TCP-IP and the factors were 
numerous, but from personal experience and in no particular order:

- OSI was primarily in a world of mainframes and large servers with dumb 
terminals but in the meantime the IBM PC had won the desktop.
- There was no easy to install OSI software for X.3 / X.25 / X.400 on a 
PC. In order to get anything working from your PC you'd need to run as a 
dumb terminal (VT55 or VT100 emulator) connecting to the serial port of 
your PC that was connected to a modem, over to some mainframe. And it 
all sucked. (sorry there is no other word I can use for it)
- To do file transfer you'd need to run Kermit, or Xmodem/Ymodel/Zmodem 
- and trust me, this was a pain to get working too

By the early 90s came a few things that changed the game for us:

- Phil Karn's KA9Q NOS that made life ever easier to run a TCP-IP stack
- Novell Netware that would manage to just about... work... with 3COM 
ethernet cards... and all the cheap NE1000, NE2000 knock-offs (with 
frequent freezes, and later, when Win3 got used instead of DOS 3.3, blue 
screens of death!)
- Trumpet Winsock - https://en.wikipedia.org/wiki/Trumpet_Winsock
- Cheap/free software like PCRoute that could turn a cheap 286 PC with 
the above cards into a router between the thick Ethernet backbone 
(10base5) and our local thin Ethernet (10base2) and... to our amazement, 
10baseT which was a godsend compared to 10base2. (anyone who'd been 
running 10base2 cabling would know what I mean)

It was cheap and it worked. Telnet and FTP mostly worked out of the box. 
>From home, the above softwares would let you use your PC with a modem 
and run SLIP/PPP and it worked, and then when US Robotics got the modem 
to work at 56Kbit/s there was no way back.

I made many demos to companies in the City in 1993-4 using a variety of 
set-ups as above and the majority were still stuck in Token Ring + 
proprietary protocols because (a) "TCP-IP and Internet was not proven 
technology" and (b) what do you mean, this is not made by DEC or IBM? :-)

O.



-- 
Internet-history mailing list
Internet-history at elists.isoc.org
https://elists.isoc.org/mailman/listinfo/internet-history