The 1997 ETI Study

The Effect of Internet Use on the Nation's Telephone Network was released on January 22, 1997. It tested the claim that dial-up traffic was breaking the phone system, and the claim did not hold.

Why a coalition buys a study

By late 1996 the argument over access charges had reached the stage where both sides were asserting facts about the physical network and neither side's assertions were easy for a regulator to check. The local telephone companies said dial-up traffic was congesting their switches. The technology industry said it was not. Someone had to put numbers on it, and the numbers that mattered were locked inside carrier engineering data that only the carriers had submitted, in the form they chose to submit it.

The Internet Access Coalition's answer was to hire an outside analyst with a track record of taking carrier filings apart. Economics and Technology, Inc. of Boston had been doing exactly that kind of work since 1972, chiefly for state commissions, consumer advocates and competitive carriers. Lee L. Selwyn, ETI's founder and president, and his coauthor Joseph W. Laszlo produced a report of more than fifty pages across five chapters and roughly a hundred footnotes. It was released on January 22, 1997, and published in full on this domain, then the coalition's website.

Five filings in the record

The report was not a general essay on internet growth. It was a rebuttal, and it named what it was rebutting. Chapter one identifies five documents: a Bell Atlantic report of June 28, 1996; Pacific Bell's ESP impact study of July 2, 1996; a NYNEX letter of July 10, 1996; a U S West ESP network study of October 1, 1996; and a 1996 Bellcore analysis by Amir Atai and James Gordon. Chapter four then works through the carrier filings one at a time rather than treating them as a single position, which is the part of the report that has aged best as a piece of technical criticism.

The public version of the claim was less careful than the filings. Pacific Telesis chief executive Philip J. Quigley described a coming network meltdown to the San Francisco Examiner in October 1996. That framing is what put the issue in front of general readers, and it is also what made the argument vulnerable, because a meltdown is a falsifiable thing to promise.

"meltdown in the network"

Philip J. Quigley, chief executive of Pacific Telesis, San Francisco Examiner, October 4, 1996

Counting switches, counting revenue

The first line of attack was the sample. The carrier studies rested on measurements taken at 127 central offices, out of roughly 5,200 in the filing companies' own territories and 23,686 switches nationwide. That is about 2.4 percent of their own footprint, and the offices had not been drawn at random. They were selected because they served internet service providers, which is to say they were the sites where the effect was most likely to appear. Findings from a sample chosen for its extremes describe the extremes, not the network.

The second line of attack was accounting. The carrier studies measured the terminating end of a dial-up call, where the internet provider's lines sat, and treated that as the cost of internet traffic. ETI argued that this was half a ledger. Every one of those calls also had an originating end, a residential line that somebody was paying for, and in a great many cases a second residential line bought specifically so the household could stay online without tying up the telephone. Counting the cost at one end while ignoring the revenue at the other is what produced a picture of the internet as a burden.

Where congestion can and cannot live

Figure 1 in the report is a schematic of a dial-up call broken into eight elements, and it does more argumentative work than any other page. Elements one through six trace the call through the telephone network: the subscriber line from the home computer and modem, the originating end office switch, the interoffice trunks, the tandem switch, the terminating end office switch, and the access lines running into the internet service provider. Element seven is the provider's own modem bank and servers. Element eight is the internet itself.

The point of the breakdown is jurisdictional as much as technical. Elements one through six belong to the local exchange carrier. Elements seven and eight do not. A user who experienced a busy signal, a slow page or a dropped session in 1997 was very likely running into an overloaded modem pool at their provider or into congestion somewhere out on the packet networks, and neither of those is a reason to charge that provider by the minute for the use of a telephone switch. The original graphic did not survive the early web crawlers, so the version shown here has been redrawn from the surviving caption text.

"no deterioration of service in our network"

Pacific Bell, comments to the California Public Utilities Commission, November 14, 1996

Concentration, not the switch fabric

The technical chapters narrow the problem considerably. In a modern Class 5 switch of the period, the Nortel DMS-100 being the worked example, the shared resource that is genuinely sized below the number of lines it serves is the line concentration equipment sitting between subscriber loops and the switching fabric. That is where a long modem call can actually crowd out a neighbor's dial tone. The switch processor and the trunk side are engineered differently, and the report argues that treating the whole switch as a scarce resource overstates the exposure by a wide margin.

It also points out that the fix was already in commercial use. Internet providers taking service on the trunk side, over ISDN primary rate interfaces rather than banks of ordinary business lines, bypass the concentration bottleneck entirely. By the report's account roughly half of Bell Atlantic's internet provider customers were already connected that way. And the cost figures were not large in carrier terms: Bell Atlantic's own worst-case estimate for the switch upgrades at issue came to about 2 million dollars, against roughly 409 million dollars of annual switching investment reported in its 1995 ARMIS filing.

A factor of six

The revenue side of the report is the part that has been cited most often. Additional-line penetration in American households had held at roughly 2.5 to 3 percent through the 1980s and reached 14.7 percent by 1995. About six million residential lines were being used primarily for online access in 1995, generating around 1.4 billion dollars of revenue in that year alone, and 3.5 billion dollars cumulatively since 1990. A small discrepancy is worth flagging for anyone working from the original files: the study's table of contents page gives the cumulative figure as 3.6 billion dollars.

Set that against the cost the carriers were asking to recover. Bellcore's estimate of network reinforcement for internet traffic came to about 35 million dollars a year per regional company, or roughly 245 million dollars nationally. The revenue already flowing from internet-driven line growth exceeded that by a factor of about six. And most of it was earned on plant that already existed: Bell Atlantic chief executive Raymond F. Smith told a Merrill Lynch audience in March 1996 that second-line sales had risen more than 50 percent in 1995 and had been provisioned out of idle capacity in existing facilities.

How the claim came apart

The meltdown framing had already begun to soften before the study appeared. A Pacific Bell spokesman, John Britton, walked the characterization back in the San Jose Mercury News in November 1996, and the company told the California Public Utilities Commission that same month that it was seeing no deterioration of service in its network. Once the strong version of the claim was withdrawn, what remained was a localized capacity question at particular switches in particular evening hours, which is a network planning matter rather than a case for restructuring how an entire industry is billed.

This is the useful part of the episode. The carriers had a legitimate engineering observation about holding times on a network built for short calls. What they did not have was measurement broad enough to support the remedy they were asking for. The study did not win by denying the phenomenon. It won by insisting the phenomenon be sized, and by pointing out that once sized it was small, local, already profitable, and fixable with equipment the carriers could buy.

What survives and where

The final chapter looks forward rather than back. It treats the arrival of 56 kilobit modem technology, announced by U.S. Robotics on January 9, 1997, as the practical ceiling of what the analog network could deliver, and argues that the real trajectory runs through packet-switched architecture, unbundled network elements and competitive entry. Read in 2026 that is a fairly accurate outline of the two decades that followed, which is part of why the report keeps turning up in citations long after the docket it was written for closed.

The full HTML edition the coalition published in 1997 survives at its original addresses on this domain: an executive summary as the coalition framed it, ETI's own summary, the table of contents, and five chapters with their footnotes. Appendix A was distributed as a separate PDF and does not appear in the archived record. One bibliographic wrinkle: ETI's own publications list dates the report to July 22, 1997, while the coalition's site, which published it, gives January 22, 1997. The primary source is used here. The text itself remains copyright 1997 Economics and Technology, Inc., preserved in archival form at the domain that first carried it.