How the Bell System Missed the Internet 8

By

“Epilogue”

In this series I have been talking about how in the 1970s AT&T’s Billy Oliver had the vision to build a packet switched data network. Billy was the Steve Jobs of the Bell System. He was a visionary and a very good engineer. He saw how data traffic was consuming the POTS network and he new that packet switching was the future.

His vision was simple: Data customers would connect to a second, parallel network optimized for data. A channel could transmit only 9600 bits per second in analog mode, but could transmit 64000 bits per second in digital mode, almost an 8-fold increase in capacity. Plus, most users only utilized 25% of their circuits, so really a 20~30-fold increase in capacity was possible.

The ingenious design meant that a customer on the ACS (Advanced Communications System) system would have a digital telephone number (DTN) and could call any other customer just like a regular phone call. If both were ACS customers then the connection would be purely packet switched. If one of the customers was a POTS customer then the call would be switched as much as possible through the ACS packet network would would then gateway through a modem into the POTS network. Thus, anybody could use the new network and migration would be a natural thing. Customers would pay by the packet, rather than by the minute.

Early on, AT&T had looked at the ARPANet system being designed by BBN. The problem with ARPANet was that it didn’t scale well. It was very processor intensive and required mini-computers and dedicated ports. It simply was not able to cost-effectively thousands or hundreds of thousands of customers…much less tens of millions of customers.

But two things happened that changed the world:

  1. Bell Labs flubbed up on the design of ACS and after a billion dollars they could not build it. Internal politics refused to acknowledge problems and allow a redesign, and so the company threw hundreds of millions of dollars a year into trying to force an architecture that was flawed at the core.
  2. The military continued to pour money into ARPANet, and BBN continued to reduce the complexity of packet switching. First, they built massively parallel Butterfly computer (http://en.wikipedia.org/wiki/BBN_Butterfly) which dramatically increased packet switching capacity. Then, packet switching became so mainstream that it was implemented in silicon directly.

The Bell System’s vision of migrating people naturally over to a digital network was elegant from an evolutionary standpoint. But in the end, the mighty Bell Labs couldn’t make it work. Meanwhile, packet switching became cheaper by orders of magnitude. With companies starting to adopt ethernet for data between computers in the late 1980s, packet switching got reduced to silicon and switching became trivially cheap. The BBN TCP/IP protocol v4 was able to be layered on top of ethernet and as it was a government funded project the protocol was free.

I personally don’t think that the BBN Internet was that profound, although in the end it did prevail. Any of the emerging packet network systems could have been the victor. The Bell System design was too early. Bell was trying to build a system in the late 1970s and early 1980s that wouldn’t really be feasible for another decade. After that decade had passed, Billy Oliver (Bell’s version of Steve Jobs) had retired and the Bell System had been broken into smithereens. There simply was no longer a central company able to fund a massive R&D project and Bell Laboratories had all but ceased to exist…the Holmdel lab was closing down permanently.

Inventions have their time. They are like a river flowing. Gravity propels the water down. The water may stumble upon a boulder or a dam, but it will reach its destination. Most inventions are inevitable once the pieces fall into place. This is why so many patents are concurrently filed for the same thing…the pieces have simply fallen into place.

The Internet was inevitable. Yes, BBN’s ARPANet was the winning design. But the inertia of data communications commanded that some packet switching system would prevail. Really, it didn’t matter which technology it ended up being.

There is a lesson from this story, too. That lesson relates to engineering management. For some things the time isn’t right. For some things the technical design isn’t right. Usually, on a doomed project, adding more people and spending more money does not fix the project, it only drowns the entire company. And, even esteemed engineering organizations like Bell Laboratories are political and mat not quite tell the truth…

Part 1

Colin Berkshire