How the Bell System Missed the Internet 1

by Colin Berkshire

Colin here. One of the more fascinating twists and turns in corporate history is how the Bell System missed developing the internet. What follows is the story of ACS in 8 parts (in 8 days). It could have owned the internet, except for NIH (Not Invented Here.)

In the mid 1970s I worked at Bell Laboratories in Holmdel New Jersey on a project called “Advanced Communications System” (or ACS.) This was a project destined to become what the internet is today. The Bell System sunk a billion dollars into its development…it was the largest privately financed engineering project in the history of mankind. The Bell System knew how important data was to become and it set out to build a vast data network.

I joined the project mid-way in a special capacity. Officially I worked at Bell Labs to coordinate the acceptance testing of the system before its roll-out across America. But unofficially I was to report back to AT&T Long Lines, which was the division funding the entire thing. Officially I reported to a manager at Bell Laboratories, but unofficially I reported to a senior vice president at AT&T. And, this was no ordinary vice president; it was Billy Oliver.

Billy Oliver was the Steve Jobs of the Bell System He is the man who invented and rolled out 800 service as an automated way to eliminate Zenith and Enterprise calling. He was the man who championed the 4ESS system. He funded the 3B computer that became the processing backbone of the Bell System and which was specifically designed to run Unix. Billy Oliver really was Steve Jobs…Bell’s all-powerful visionary. He was good at politics, trusted, a very capable engineer and comfortable with multi-billion dollar budgets within the world’s largest corporation.

(See http://www.ieeeghn.org/wiki/index.php/Billy_B._Oliver. See 1989 Alexander Graham Bell Medal: http://www.ieee.org/documents/bell_rl.pdf See Development History of 4ESS: http://en.wikipedia.org/wiki/4ESS_switch)

In the early 1970s Billy Oliver noticed how data was increasing as the fastest growing source of call traffic. The percentage growth of phone calls doing data was unbelievable. And, the Bell System was getting strangled by corporations ordering private lines crisscrossing the United States. AT&T hated private lines because they consumed a dedicated channel and they required lots of engineering time to install. Every time a new dedicated point-to-point data circuit was installed, a channel was removed from the shared network. Dedicated channels bypassed the company’s automated circuit testing systems, and when they went down they were difficult to debug. Worse, we couldn’t reliably deliver consistent channel quality, especially to smaller cities. The analog network was just not designed for data.

Back then you could only push 9600 bits per second through a voice circuit. But if you used that same circuit is an all-digital mode it could pass 64000 bits per second…a 6-fold increase in capacity. We discovered that customers only utilized about 25% of the circuit bandwidth and thus they wasted 75%.
The math was compelling: a 6x improvement in throughput multiplied by a 4x improvement in bandwidth utilization was a 24x improvement. Adding in the ability to utilize circuits for residential traffic after hours this meant about a 36-fold economic benefit from a switched data network.

More importantly was getting data off the PSTN public network. It was strangling the network. Voice circuits are designed to carry roughly 6 minutes of traffic per peak hour. But a data circuit might be dialed up and be left up all day. Instead of generating 30 minutes of traffic a day a data circuit would generate 480 minutes a day…sometimes even 1,440 minutes because some companies just left them up all the time. In areas where there was flat-rate service it was an economic catastrophe. A Central Office like the 1ESS could handle 100,000 calls an hour if they were 6 minute calls. But if the calls were nailed up (which was the term for a data connection that would just be left up) the network junctors would block at very high rates and call capacity would decrease 10-fold or more. Data was devouring the public phone network.

Billy Oliver very much understood these numbers and an imperrative effort was made to build a high capacity data network that would parallel the analog voice network. This problem was so urgent that the largest engineering budget ever in the Bell System (and probably of any company in history) was authorized.

This project was to be called “Advanced Communications System” or ACS. It was to be developed at Bell Laboratories in Holmdel New Jersey. (Which was the birthplace of the Morris system and the 1ESS, by the way.) Thus, the laboratory in 1B5 was born.

I was flown up to meet with some professors at MIT and Harvard and then to meet with folks at BBN. MIT and Harvard had connections to the ARPANet which was being developed by BBN. So it was a parallel effort that perhaps we could tap into.

Users at Harvard and MIT were enthusiastic about their ARPANet connections. I was impressed that the AP newswire could be received and articles meeting pre-selected keywords could be saved into a user’s read file. Users were also exchanging electronic mail messages…between disparate systems. Within the Bell System some people had email, but only similar, compatible systems could talk and addressing was a nightmare. (To address an email message you routed it like a step-by-step switch, specifying each hop, so your email address might be billy!rudolph!bedmin!attll!btl. This was called “bang addressing and used the UUCP protocol.)

But at Harvard and MIT people could send emails to anybody, anywhere, on any system. You didn’t need a connection directly to the recipient’s computer system. Messages would pass through cooperating organizations and circuits. A path would be automatically found. At that time, the ARPANet ran at speeds of 9600 bits per second between the few switches, but there were plans to upgrade to 64 kilobit circuits once the BBN butterfly was completed.

Next…The problem with ARPANet…It doesn’t scale…