A Practical Navigator for the Internet Economy

Qwest Plants Fiber and Leverages Result

Exploits New Technology to Leapfrog Older Telcos

Nayel Shafei Explains Bandwidth Strategy: Will Nurture Demand But Will Not Start Price War - Rumored Bandwidth Glut Meets Reality of Paying for Non Congested Service pp. 1 - 8

Discussions of bandwidth, the most critical part of Internet infrastructure, are going in contrary directions these days. While huge amounts of bandwidth are becoming 'available' what is not, as yet, at all clear is the impact of companies like Qwest and Level 3 on the marketplace. Followers of the Gilder paradigm continue to pledge that bandwidth will become like transistors, so cheap as to be almost free. But for now the price of internet leased lines increases in the face of incredible traffic growth.

As the two technical articles that follow our Qwest interview show, the current business model of the Internet seems predicated on squeezing the last drop out of available bandwidth. The predominant best-effort-of-packet-delivery business model is coming under stiff criticism for encouraging the over selling of circuits and under provisioning of networks. He who out multiplexes his competitors wins.

Charging for first class traffic by the packet miles involved in its delivery is beginning to appear as a conceptual alternative approach that will encourage providers to provision their networks adequately. We find the argument compelling but also ironic, coming as it does, in the midst of all the talk about huge new supplies of bandwidth and the assumed collapse in its price. Of course while it is likely that none of the new entrants (which also include Level 3 and IXC and some cable TV networks) intend to destroy the pricing umbrella, their combined impact could have this effect.

According to our interview with Nayel Shafei, Qwest CTO, when Phil Anshutz sold the Southern Pacific Railroad a few years ago, he did not sell his access to the right of way. This time, instead of microwave (Sprint), he is building a next generation network from fiber. Qwest' plow train lays two conduits that are each only a few inches in diameter. Each of these conduits can carry up to 960 fibers. Qwest is using only one of the two conduits, leaving the other standing empty. In the one they are using they are placing only 96 fibers. They are selling 48 and using the proceeds to finance their build out. The other 48 they will use for their own network. Saying that, from the very beginning, Qwest always intended to be a full services telecom company, Shafei touts the vast amount of bandwidth Qwest is laying. Through bandwidth trades with owners of trans-oceanic cables Qwest will expand globally.

Shafei remarks that he has gotten Sun to promise an OC 12 port on the back of its machines by July 1 and that he has met with Andy Grove to encourage Intel to work with various folk to increase the capacity of buses along the lines of the Symbolics machines of the mid 80's. When asked whether Qwest would gain market share by drastic undercutting of current pricing, he responded: No. "You see more important than breaking or destroying the current pricing umbrella is to come up with a new pricing rationale that will hold fast all the way from T1 up to OC-192." "We are experimenting now with at least three different pricing structures with various customers. One of these structures is what you might call bit buffet: you pay for what you eat."

Qwest has seriously entered the backbone and the ISP market with its acquisition of Colorado Supernet and EUnet in Europe. Nayel stated that, while one can buy Qwest backbone access now, specific pricing and details about network topology would not be available for another two months.

We asked about the deal to provision AGIS' backbone, wondering if Qwest was getting access to AGIS' peering agreements with the big five. He said that was not the case and that the deal was essentially a 20% stake in the company, plus further payments due Qwest at the end of two years. If AGIS should not make these payments, Qwest gets in return a larger share of the company. Finally "they must pay the ordinary operations and maintenance fees needed to maintain the proportional share of the network bandwidth that they are using."

Economics of Interconnection, pp. 9 - 11

On Inet-Access at the beginning of April, a debate occurred that offered evidence of the Curran hypothesis of what happens to the customers of networks that do not have adequate connectivity with the big five. In this case an up stream provider of DRAnet had such poor connectivity with Sprint at the public exchanges that DRA felt compelled to take a direct connection from Sprint to get around the problem.

In a debate with Sean Donelan and others Sean Doran made some provocative points. We are seeing in his arguments the formulation of a business model designed to replace the best effort model. The new model would reward pro-active investment in new bandwidth. Doran remarked: "Note that distance-and-volume sensitive pricing, *provided* networks do proper provisioning, erodes the need for NAPs and MAEs, too, by making it cost-effective to exchange local-traffic through a transit provider. "

[In a packet charging business model,] "each packet would be a very small fraction of a dollar, say. Make a ceiling, which would be something like the normal flat-rate price plus or minus some epsilon, just to reassure your customers that they won't lose their shirts. Consider bumping up your flat fees a little if you feel you need to encourage your customers to save money by making it easier for you to properly provision your network."

"In that vein, make sure that you don't lose paid-for-per-packet packets instead of flat-rate packets, in the event of congestion[*]. Then point customers at various tools and ideas with respect to congestion sensitive traffic admission control, packet shaping, or whatever you want to call it. . . . . Diff-serv is useful in bootstrapping oneself up towards a properly provisioned network: mark packets "flat-rate" versus "unit-charged", and treat the latter as least droppable after network control and management traffic."

Diff-serv Takes Shape, pp. 12 -17

The Diff-serv Working Group went into high gear just before the LA IETF. Much discussion on the diff-serv list focused on the Nichols draft and its definition of PHB (per hop behavior). A significant portion of the debate focused on the latitude allowed for the specification of per hop behavior. Should the standard allow the backbone provider wiggle room in its implementation because some backbone providers claim that they need something that works now and cannot predict what future requirements might be? The suggestion is made that diff-serv can be an environment in which developers can refine the behavior of MLPS when it crosses administrative boundaries.

An interesting exchange from the list discussion "So, is the goal of working out a charging mechanism so that providers can make money deploying support for end-to-end services for traffic traversing their networks in scope for this WG?"

"No. But it should certainly be a goal not to preclude such a mechanism. Because - if you do - why bother offering premium services - you'll never be able to bill anyone for them. "

Symptoms of Inadequate Connectivity, pp. 18

We asked Sean Doran to describe what a corporate customer would experience if his upstream provider had inadequate connectivity to MCI/WorldCom. Sean does this and then goes on to say that he is not sure what would actually happen, given the merger's approval.

Internet Telephone Strategy of Telecom Finland, pp. 19 - 22

An interview with Hannu Tuomisaari, Manager of Intranet Communications Services with Telecom Finland shows how this creative PTT is using internet telephony to provide useful business intranet services and compete with private telcos that dominate Finland's largest cities.

Attacks on ARIN, Editorial, p. 24

We are alarmed at recent attacks on ARIN where false accusations have been sent to public lists and NTIA. The ARIN Board and Advisory Council need to decisively rebut them and ensure that ARIN doesn't come into play in the NTIA rulemaking process.