The executive summary that answered the meltdown scare

In January 1997 two economists at Economics and Technology, Inc. published a study arguing that the phone companies had the internet problem backwards. Their executive summary is one of the clearest statements of what was actually at stake.

What the phone companies told Washington

Through the summer and autumn of 1996 a series of filings landed at the Federal Communications Commission from the largest local phone companies. Bell Atlantic filed a report on June 28. Pacific Bell followed with an ESP impact study on July 2, NYNEX with a letter on July 10, and U S West with a network study on October 1. Bellcore, the research arm shared by the regional Bell companies, produced its own analysis by Amir Atai and James Gordon. Together they made a single argument: calls to internet service providers were tying up switches built for short voice conversations, and the companies were carrying the cost without being paid for it.

The rhetoric outran the filings. Philip J. Quigley, chief executive of Pacific Telesis, warned the San Francisco Examiner in October 1996 that the network faced a meltdown. The remedy the industry wanted was straightforward and expensive: strip internet providers of the exemption that had shielded them from interstate access charges since the FCC's 1983 MTS and WATS Market Structure decision, and start billing them by the minute. Because providers would pass those charges on, the practical effect would have been a per-minute meter on a service that consumers had only just learned to buy at a flat monthly rate.

Who wrote the study, and for whom

The Internet Access Coalition, a group of twenty-four technology companies and trade associations including America Online, Apple, IBM, Intel, Microsoft, Netscape, Oracle and Sun Microsystems, hired Economics and Technology, Inc. of Boston to test the claims. ETI was not a lobbying shop. Its president, Lee L. Selwyn, held a doctorate from the MIT Sloan School and had testified on telecommunications economics before roughly forty state commissions, the FCC, Canada's CRTC, the United Kingdom's Oftel and committees of both houses of Congress. He wrote the study with his colleague Joseph W. Laszlo.

The result, titled The Effect of Internet Use on the Nation's Telephone Network, ran past fifty pages and was released on January 22, 1997, two months before comments were due in the FCC's access reform proceeding, CC Docket No. 96-262, and its companion inquiry into data-friendly connections, CC Docket No. 96-263. The executive summary preserved here is ETI's own condensation of that argument. One small archival note: ETI's later publications list dates the report July 22, 1997, while the coalition's own site says January 22. The coalition's date is the primary source and the one used throughout this archive.

"Any predictions that Internet traffic will soon result in a meltdown of the network are greatly exaggerated."

Selwyn and Laszlo, 1997

Congestion was local, not systemic

The first conclusion was about scale. There were 23,686 central office switches in the United States. The Bell studies rested on observations at 127 switching entities that happened to serve internet providers, roughly two and a half percent of the offices in the filing companies' own territories and a rounding error nationally. Selwyn and Laszlo did not dispute that those particular offices had strained. Their objection was to the leap from a handful of hot spots to a claim about the whole network, and they attributed most of the trouble to ordinary planning and engineering failures that existing equipment configurations could fix cheaply.

They also separated the network into parts that the Bell filings had blurred together. A congested line into a single provider, a provider's own overloaded modem bank, and congestion out on the internet itself are three different problems, and none of them degrades service for the person next door making a voice call. The study's network diagram, which the archive's crawler never captured, existed to make exactly that point. There was also an awkward fact for the incumbents: several of them were at that moment selling unlimited flat-rate internet access themselves, which is not the behavior of a company that believes such traffic is breaking its network.

The traffic paid for itself six times over

The second conclusion was financial, and it was the one that did the most damage. Households that went online bought second phone lines, in large numbers. Additional-line penetration rose from between two and three percent before 1990 to 14.7 percent by 1995. ETI calculated that about six million residential lines were in use principally for online access in 1995, generating $1.4 billion in revenue that year alone, and more than $3.5 billion cumulatively from 1990 through 1995. A table of contents page on the coalition's site cites $3.6 billion since 1990, a small discrepancy worth flagging but not one that changes the argument.

Set that against what the industry said the traffic would cost. Bellcore estimated network reinforcement at roughly $35 million per year per Bell company, or $245 million nationally. The revenue from online-driven second lines exceeded that figure by a factor of six. Individual numbers pointed the same way: Bell Atlantic's worst-case switch upgrade was about $2 million, against annual switch spending of $409 million reported to the FCC for 1995, and its own chief executive Raymond F. Smith had told a Merrill Lynch audience in March 1996 that second-line sales were up more than fifty percent, supplied from idle capacity already in the ground. Because heavy internet use fell outside the voice busy hour, it was filling capacity that would otherwise have sat unused, which lowers average cost rather than raising it.

"No deterioration of service in our network."

Pacific Bell, comments to the California PUC, November 14, 1996

Build better networks, do not meter the old one

The third conclusion refused the false choice the debate had settled into. ETI agreed that a circuit-switched network designed for three-minute voice calls was a poor long-term host for data, and said so plainly. Where it parted company with the Bells was on the remedy. A per-minute charge on internet traffic would raise money inside the old architecture without producing any of the new architecture, taxing the symptom while leaving the cause in place.

The alternative the study set out was investment in packet-switched and data-friendly facilities, and the digital subscriber line technologies then arriving, ADSL and HDSL, which the coalition argued could lift household speeds by a factor of a hundred over the 28.8 and 33.6 kilobit modems of the day. US Robotics had announced its 56k X2 technology on January 9, 1997, days before the study appeared, and that was close to the ceiling of what copper voice-grade service could deliver. Getting past it required competition and unbundling, so that incumbents could not simply decline to build and could not block the entrants who would.

Why the argument still matters

The immediate fight went the coalition's way. The FCC's tentative conclusion, reached in the notice it released on Christmas Eve 1996, was that internet service providers should not pay interstate access charges, and that exemption held. Flat-rate internet access survived, which is the reason the late 1990s were a dialup boom rather than a metered trickle. The coalition's own arithmetic had put the stakes at up to sixty cents an hour on top of a $19 monthly plan, twelve extra dollars a month for a modest twenty hours of use, at a moment when most households were deciding whether the internet was worth paying for at all.

What makes the study worth revisiting is that the second half of its argument came true as well. The circuit-switched network really was the wrong long-run answer, and the transition ETI called for did happen, through DSL and then cable and fiber. Selwyn's firm followed the thread directly, publishing Building A Broadband America and Bringing Broadband to Rural America in 1999 and broadband policy work into the following decade. The structure of the 1997 dispute has also proved durable. An incumbent argues that a new class of traffic is straining its network and should pay extra to ride it; the counterargument is that the traffic is already paying, that the strain is narrower than claimed, and that a metering scheme substitutes for the investment everyone actually needs. That exchange has recurred in nearly every network policy fight since.