Five filings, examined one at a time

Chapter 4 took the five carrier filings separately. Economics and Technology, Inc. read Pacific Telesis, US West, NYNEX, Bell Atlantic and Bellcore on their own terms, then asked in each case what the company had actually measured.

A two page filing and one switch

Pacific Bell filed its ESP Impact Study on July 2, 1996: a two page summary drawn from measurements inside central offices serving enhanced service providers. The underlying observation was sound and ETI did not dispute it, since calls to an online service run long and provider lines carry more traffic than an office average. But Pacific examined roughly 1.8 percent of its lines and fewer than 5.2 percent of the providers in its territory, over two weeks, with no comparison group of ordinary customers.

Every reading also came from the terminating end, where the load shows up and the money does not. Pacific put provider revenue near 26 million dollars by multiplying an estimated 110,000 access lines by a 20 dollar monthly business rate, counting recurring charges only and nothing from the calling side: measured business usage, flat rate residential service, second lines, ISDN. Its 2.6 million dollars of early 1996 capital spending, and 11 million forecast for ISDN Primary Rate, were never separated from routine upgrades.

The public case outran the filing. CEO Philip J. Quigley described an approaching network meltdown in the San Francisco Examiner on October 4, 1996. By November 8 spokesman John Britton was giving the San Jose Mercury News a far narrower account, and on November 14 the company told the California Public Utilities Commission that service had not deteriorated.

Sixty four hunt groups in four states

The U S West ESP Network Study, dated October 1, 1996, put two pages of conclusions above ten pages of charts, built from data on 64 provider hunt groups in four states. Its central finding was that provider usage patterns differ from those of other end users. ETI accepted that, then noted how little followed. The study named no specific instance of trouble caused by data traffic and quantified no investment made to relieve it.

Timing was the sharper claim. Provider traffic peaked around 10 pm while the company's central office peak stayed near 4 pm, and U S West treated the gap as future reengineering work. ETI read the numbers the other way: evening traffic uses switching capacity that would otherwise sit idle, and mixing loads that peak at different hours raises throughput and lowers average cost per minute. On U S West's own attachments, no studied office had yet shifted its busy hour to the late evening.

Growth projections arrived without stated method or source and assumed all future demand would be analog modems on the circuit switched network, even as the company's own digital PBX tariffs in Minnesota and Oregon fell. And the cost per line chart treated terminating usage as the cost causing event, against the sent paid structure of U S West's tariffs.

"we were able to provision new lines and services from idle capacity in an existing plant"

Raymond F. Smith, CEO of Bell Atlantic, Merrill Lynch Telecommunications CEO Conference, March 19, 1996

Blocking at the switch, or at the ISP

NYNEX wrote to the FCC on July 10, 1996 with six charts of provider hunt groups. Its cover letter reported about 200 companies using analog dial up connections in its territory, with those lines growing roughly 10 percent a month. Here the concern was concrete: calls really were failing to complete, and blocked attempts appear plainly in the data.

What the charts could not say was where the blocking happened. They showed only calls placed to the selected providers, with nothing about other traffic through the same switches. A Hempstead office handling several hundred busy hour attempts blocked most of them, and the provider there had configured a hunt group of 22 lines. That is an undersized hunt group, not a congested switch. A White Plains office in the same set showed heavy usage and no blocking.

Pricing shaped it. New York Telephone's tariff carried a substantial surcharge for PRI ISDN trunk side connection, pushing providers toward cheaper line side business lines, the very arrangement whose behavior the filing described. On the revenue side, the 1.26 million second residential lines NYNEX reported for 1995 worked out to about 33.3 million dollars, against roughly 28.9 million the year before, a stream growing 9.3 percent a year.

The deepest filing and its blind spot

Of the five, ETI rated Bell Atlantic's June 28, 1996 submission the most substantial, and its central data point cut hardest against the industry case. Of 4,887 provider circuits examined, half used PRI ISDN trunk side connections, a share it called representative. Trunk side connections bypass the line concentration module and do not block at the terminating switch, so the filing's line unit limit of roughly 65 subscriber lines applied only to the analog half, and with PRI priced close to basic business service that was the shrinking half.

Bell Atlantic again counted revenue from one end, assuming 17 dollars per provider line per month, about 8.2 million dollars for 1996, and leaving out measured usage on calling lines, second lines, ISDN and separately tariffed features. Its own chief executive had described the other side of the ledger at a Merrill Lynch conference on March 19, 1996: secondary line sales rose more than 50 percent in 1995 on internet and telecommuting demand.

Scale matters too. Elsewhere in the study, the worst case cost of upgrading a switch for heavy data traffic came to about 2 million dollars, against the 409 million a year Bell Atlantic reported spending on switching in its 1995 ARMIS filing. The company also set its 2 cent access charge beside the 0.09 cents per minute it attributed to providers, where ETI held the benchmark should be forward looking incremental cost.

"no deterioration of service in our network"

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

Right about the decade, thin on the year

The last document was not a carrier filing but a 1996 technical study by Amir Atai and James Gordon of Bell Communications Research in Red Bank, New Jersey, widely cited by the companies. Its long range conclusion was that circuit switching handles data badly and that traffic would have to migrate in stages toward packet networks. ETI agreed outright, accepting the core engineering judgment of the paper it was rebutting.

The near term case was another matter. Bellcore modeled what reinforcing the circuit switched network would cost, arriving at roughly 35 million dollars a year per operating company and about 245 million nationally. No field measurement of congestion actually occurring accompanied the model. Long duration calls change a switch's behavior only where they are a large share of traffic at a given point, and across the network they were not.

Bellcore also asserted that second line revenue was unlikely to offset capital spending without supporting the claim, just as the operating companies were crediting second line demand for earnings growth. And apart from cable modems it assumed the fix would come from incumbents rebuilding their own networks, not from competitors building data friendly capacity under unbundling and collocation. Its own note that growing hunt groups would push providers toward trunk and PRI interfaces removed most of the switching problems it described.

What the five had in common

Read together, the filings repeat three moves. Each measured at the end of the call where providers terminate rather than where subscribers originate. Each counted revenue only from the providers, treating the calling customer's line, second line and ISDN payments as though they belonged to some other business. And each sampled offices chosen because providers were in them, with no control group.

That is a criticism of study design, not of the engineers. The concerns were real in places. Hunt groups were undersized, line side concentration does have limits, and evening peaks eventually require capacity. But each has a remedy in provisioning and tariff design, and none was shown to be a network wide condition. The gap between the rhetoric and the filings was wider still, and Pacific Telesis closed that one itself within six weeks.

The finding Chapter 4 established was narrow, and it was the one that mattered at the FCC: not that the network would be fine forever, but that none of the five submissions had measured the thing it was cited to prove. Written by Lee L. Selwyn and Joseph W. Laszlo of Economics and Technology, Inc. for the Internet Access Coalition, 1997.