The Network Is the Market: Financing Internet Bandwidth
By John du Pre Gauntt
The 15-day strike in August 1997 by workers at United Parcel Service (UPS) holds valuable lessons for the Internet. Within minutes of the walkout, UPS warehouses began filling with packages that couldn't be sent on to their final destination. During the strike, the company moved about 500,000 parcels per dayconsiderably fewer than the prestrike daily average of 12 million. In addition, many of the packages were sensitive to delaywith sensitivity ranging from the failed overnight delivery promised by mail-order retailers, to human skin graft material that had to be rushed to surgeons.
Compare that situation with the events of July 17, 1997, when huge swaths of the global Internet suddenly became inaccessible. In that case, the Internet's domain name system (DNS) that translates complex routing numbers into user-friendly names such as economist.com failed because of a software glitch compounded by human error. Again, the core of an immensely influential system depended on the fortunes of a single company: Network Solutions Inc. (NSI) of Herndon, Virginia.
The UPS strike and the NSI fiasco provide a glimpse of just how dependent modern societies have become on the smooth functioning of networks. Such dependency is, for the most part, invisible. However, it is when these networks fail, become congested, or get disrupted that society learns how much of its well-being rides on so little.
But there are certain important differences between the way people perceive their dependency on UPS and people's growing dependency on the Internet. People can see a correlation between the performance of the UPS network and the price structure that supports it. For its part, notwithstanding strikes, UPS knows its business is delivering parcelsso it knows how much that should cost. As such, there is a feedback system that informs users what their actions are going to cost and also tells UPS where it should invest its resources.
Likewise, people have been educated about the Internet but in a different manner. They are being fed a seductive idea by technologists and policy makers that says technical advance and market competition will ensure that information is free and that the future lies in an age when bandwidth flows like water and becomes as easy to access as electricity.
And as with water or electricity, all an observer need do is survey existing bandwidth markets to uncover startling price differentials and areas where service simply does not exist. In the meantime, informed observers estimate that the demand for bandwidth has been doubling every three to four months; in the past 20 years, bandwidth demand has increased 1 million times; fax transmissions make up fully half of what is counted as voice traffic; and in a few years, given certain trends, 99 percent of all traffic will be carried on Internet protocol networks.1
But for the price of bandwidth, it remains the situation that the rest of the information economy must wait as most of the main Internet infrastructure ownersincumbent telecommunications companiestry to work out ways they can migrate away from a business model that must distort per-bit prices for bandwidth in order to preserve narrowband revenue streams while making broadband service attractive to large users.
Yet there should be a point where bandwidth can be dynamically unbundled from the network without recourse to the building or leasing of dedicated infrastructure. While engineers look to protocol solutions for discriminating among classes of traffic on the Internet, work on pricing models that are both robust and simple tends to concentrate on either theoretically elegant though difficult auction systems or else prereservation schemes that add complexity to already crowded header fields on Internet packets.
The move toward traffic discrimination and bandwidth prereservation illustrates a larger shift in Internet engineering toward performance over ubiquity. Whether or not this threatens the homogeneity of the Internetin effect, creating an Internet for the wealthy and an Internet "for the rest of us"is not known. Neither is there a price mechanism that indicates the cost that prereserving bandwidth imposes on the network as a whole.
Worst of all, most pricing scenarios operate at the level of national or regional markets. There is no effective global benchmark for comparing bandwidth prices in the same fashion as, for example, Brent crude is invoked in global petroleum markets. In essence, if the Internet is to become the Global Information Infrastructure (GII), it would seem plausible that the price of bandwidth should evolve along lines similar to the prices of other global commodities such as energy, primary foodstuffs, and textiles. This implies a new type of participation by markets, which have hitherto been limited to investing in network companies as opposed to network resources.
Bandwidth Pricing Gets Serious
If one remembers the role of fossil fuel energy in the well-being of industrial civilization, one can imagine future perceptions of telecommunications platforms in the prosperity of information civilization. As more production and consumption decisions become predicated on the speed and reliability of interactive telecommunications infrastructure, the shift toward widespread commercial transactions over the public networks could redefine bandwidth as a volatile resource.
Rapid price swings are possible given that the communications costs of transacting in electronic markets made possible by the Internet are as yet unknown because the majority of present telecommunications firms' charging formulas are connection oriented. Each call has a setup phase during which a connection is established and maintained for the length of the call.2 Conceptually speaking, this model assumes no one else can use the circuit. Thus, only a single accounting record is needed regardless of the session's duration.
It's also useless for an electronic market. In a packet-switched environment, a communications session is divided into discrete packets that traverse the network separately.3 Accounting for server usage on the Web requires a separate record for every hit, which adds up rapidly even if a user perceives a continuous session. If telephone-style accounting were used, the equivalent of a one-minute call could generate more than 2,000 accounting records; a 10-minute call could involve accounting for over 20,000.4
As such, capacity planning on telecommunications systems has become an even more esoteric art than before. The former method of interrogating voice switches to build a model does not work for Internet traffic. Aside from the oft-heard stories of Internet connections that last for hours instead of minutes, on a more fundamental level, engineers are finding that data networks sometimes exhibit fractal, or self-similar, characteristics. Basically, this means that data traffic is not periodic and therefore will not smooth out over time. Average throughput may look acceptable when analyzed over five-minute intervals yet exhibit unacceptable variability over five-second intervals. The upshot for the short term is that network designs must be more conser-vativein other words, expensivein order to provide a consistently high-quality service.5
This is happening while the most important issue facing Internet commerce has yet to be resolved: how communications costs are to be allocated in a transaction setting. If a customer accesses the site of a seller, then the customer is paying for the communication. If that customer then decides to make a purchase of a product or service, should the final price of the item include the customer's communication costespecially for an entertainment site providing bandwidth-hungry audio or video files? Would there be a discount for a customer who made a quick decision and thereby conserved resources? or how about an extra charge for dithering over a purchase? How would network congestion in downloading the requested item be treated: as part of the overall price? or would there be scope for express delivery? What of wide-area collaborative applications such as gaming? Who pays the communications costs of returning an information-based item? Is there scope for a vendor to extend extra communications capability as part of a loyalty program?
It follows, therefore, that a transaction for even a simple item involves a complex communications dialogue that alternates between secure and insecure, priority and nonpriority, flat rate or usage sensitive, and so on. Thus, it is likely that future charging mechanisms will need to have knowledge of the application protocol being used.
Yet, historically, telecommunications prices have been based on distance rather than the nature of an application. Theoretically, the farther a signal had to travel, the higher the switching and transmission costs. That not only gave a basis for a pricing structure but also provided the means for cross-subsidizing telephony service. Whatever the past merits of charging high international and long-distance rates to keep the cost of local service affordable, the practical reality is that this regime for pricing bandwidth assumed (1) the predominance of voice telephony, (2) the willingness of corporate users to pay higher bills, and (3) isolated national marketsassumptions all of which now are unraveling.
To stave off competitive threats by nontraditional bandwidth suppliers such as electricity companies, most telecommunications operators have taken the idea of cost-based pricing to the center of the bandwidth debate. New entrants are flashing their equipment invoices to show that they "know" their costs, and telecommunications economists are furiously publishing book and journal articles on how one can perform a proper cost estimation for an ex-monopoly provider migrating to an open market.
Be that as it may, it remains the case that the cost and price of bandwidth are more a function of the alternatives rather than the actual cost of delivering it. In other words, the willingness to pay is going to be negotiated from a valuation point of view. "We have a service. They have a need. And they are willing to pay a price," says one European supplier. "But that price will be compared to what other options are available and what value customers can make using our facilities."6
Granted that that sentiment may be widespread in the present telecommunications industry, the fundamental fact is that the cost of bandwidth has entered into the overhead of almost any business that hopes to use the Internet for electronic commerce. Therefore, it seems unlikely that the future price of bandwidth will be determined solely by the infrastructure or marketing costs of network operators.
So if we accept that the transaction model for Internet commerce will be electronic networks and if we accept that the price of bandwidth in such a setting becomes a central business cost, then it is feasible that bandwidth could be capitalized and traded in a public setting.
There are two main reasons to trade bandwidth: to hedge risk and to speculate on its price. Hedgers and speculators not only create a new investment market; they also can unlock information about the real cost of communications for an electronic market. A trader with undisputed access to network resources is not going to be a network operatorjust as an investment bank that's trading petroleum futures does not own filling stations. Instead, what we're talking about are investors who are acting on behalf of non-facilities-based carriers or even infrastructure owners in order to control a resource that is useful, difficult to substitute, and demanded by all who transact in an electronic market.
Yet before a bandwidth futures pit can be realized, there are a host of issues that must be addressed. The primary one involves the degree to which it is possible not only to model network demand but also to model price changes over time and across multiple operators and markets in such a way that potential bandwidth contract buyerstelecommunications companies or notare able to base their decisions on the investment riskas opposed to the operational risksof actually delivering a network service. Should that admittedly major hurdle be overcome, there remains the additional need for exchange mechanisms, underwriting, and risk analysis.
As far as the modeling of network traffic patterns goes, certain companies have started producing maps of international switched voice traffic, and others are attempting to do the same with Internet traffic. The two main organizations producing those cognitive maps are Telegeography, at http://www.telegeography.com and Matrix Information and Directory Services (MIDS), at http://www3.mids.org.
Telegeography of Washington, D.C., is a leading publisher of reports on international telecommunications flows. It publishes statistics on the number of minutes of public switched trafficin millionsfor more than a hundred countries to indicate the top 20 telecommunications routes. It does this through its direct relationship with major telecommunications carriers that provide the raw data for Telegeography's annual reports. By aggregating the data over time, Telegeography produces maps of telecommunications traffic flows over the past decade in more than 50 major telecommuni-cations markets while tracking tariff changes for the past five years. The consultancy claims that the world's cross-border telephone traffic grew 13 percent to reach 70 billion minutes in 1996. In value terms, the global market for international minutes increased to $61 billionan 11.5 percent jump over 1995.
While Telegeography tracks the flow of international minutes, Matrix Information and Directory Services attempts to track Internet usage itself by means of a service known as the Internet Weather Report (IWR). The IWR produces maps of the global Internet by using the echo-request element of Internet control message protocol (ICMP)often represented by a user program called pingto query various Internet domains from its Austin, Texas, servers. Taking its list of Internet domains from the Network Wizards Internet Domain Survey at http://www.nw.com, the IWR is a sort of radar scan of the Internet, wherein the round-trip of a ping between Austin and a particular domain is modeled into a latency map. The size of the circles indicates the latency of a particular site, with the size ranging from small (low latency) to large (high latency). The IWR sends pings to the various domains five times and takes the average latency per node. MIDS then collects all of the average latencies for all nodes for each scan and makes a map. To see the change over time, MIDS uses Java to animate six scans for the day.
In addition to the IWR, MIDS has launched a further Internet visualization tool called Tracemap: http://mids.alexa.com/test/tracemap/. Tracemap enables a user to visualize the route that packets take from a server in San Francisco to a domain name specified by the user. Each Tracemap of a destination shows graphically and textually in milli-seconds the number of hops and the time between hops of the IP path from one host to another, as well as the time it took to get to an intended destination. As of February 1998, Tracemap can be used by anyone on the WWW as part of a beta test.
Granted the importance of traffic- or tariff-mapping services such as those supplied by Telegeography and MIDS, it is still the case that the only working spot markets for capacity exist in just three locations: 60 Hudson Street in New York, 1 Wiltshire Avenue in Los Angeles, and the Telehouse in London. All of these carrier hotels offer facilities colocation and disaster recovery service and provide ISPs with links into Internet backbone nodes. As such, Web-based contracts for international minutes or bandwidth are restricted largely to using these major nodes. This does little for improving local access to bandwidth. However, the more trading that goes on spot markets, the closer the world comes to standard prices for large-bandwidth chunks.
Not surprisingly, there have been several attempts at starting bandwidth trading exchanges. Three companies in particularArbinet http://www.arbinet.com, Band-X http://www.band-x.com, and RateXchange http://www.rateXchange.comare using the Web to enable buyers and sellers of international minutes or bandwidth capacity to browse multiple bids and offers before meeting.
Band-X and its direct copyRateXchangeshare a business model in which users register with the respective services and then are allowed to browse bid/offer prices that specify international routes, connection points, special technical data, and price. A buyer or seller who wishes to pursue a bid or offer to completion is introduced by the broker, who receives a percentage commission on the final agreed price.
Band-X took the concept of the neutral broker one step further via launch of its index of U.K. outbound traffic in September 1997 and launch of a U.S.-based index soon afterward. The Band-X indexes reflect movements in the wholesale prices of international telecommunications minutes.
The Band-X index is created on each of the top 20 routes by volume of international minutes. Launched with a base value of 100, the indexes are combined to create a country composite index, within which each individual route is weighted according to its proportion of total outgoing international traffic.
The data for the index are provided for Band-X by no fewer than five international carriers that submit their wholesale selling prices for the last week of the previous month. The weighting for the composite index is calculated according to data provided by Telegeography. The composite figures are released at the end of each month. The individual route indexes are released one month later and allow privileged access by the contributing carriers.
Arbinet plans to be a bit more ambitious. A technology vendor as well as a possible bandwidth broker, Arbinet proposes to build an overlay network of programmable switches on carrier backbones that can intelligently route traffic according to the pricing and quality rules established by the individual carriers. Operators postto an Arbinet server or central local node (CLN)their network availability and the prices they are offering at any given time. Arbinet customers who have miniature versions of a CLN running in their networks can query the main server for the least-cost route based on a call's particular requirements. The Arbinet overlay system calls for a universal switch that when connected to individual carrier networks will constitute the virtual clearing network. An individual carrier that wishes to join the clearing network publishes to the clearing network database full information on route quality, times, rates, and restrictions.
The clearing network operator (Arbinet) manages all of the clearing aspects. Arbinet believes the number of universal switch operators is unlikely to exceed a few thousand with a series of replicated servers. It is thought that the main argument for joining a clearing network is that whereas the marginal cost of carrying additional traffic is almost nil, the marginal cost of adding capacity is high. Between those factors, carriers that currently are exploiting less than half of their existing capacity have an incentive to manage their network bandwidth and costs by publishing to the clearing network the rates and times for which they want to transit and/or terminate other carriers' traffic.
The upshot of the emergence of network mapping services, along with exchange schemes such as Band-X and Arbinet, is that several key elements for managing bandwidth-based contracts exist. Although it is the case that data for the Band-X indexes are provided largely on the goodwill of telecommunications carriers and that the Arbinet system looks for radical overhaul of the present settlement system, these are still early days of bandwidth trading. The keys will lie in how these systems scale and how the price and volume information they are generating can be aggregated, analyzed, and integrated into workable risk models for investorstelco or not.
The thrust of those comments by Peter Bernstein is that we can
assemble big pieces of infor-mation and we can assemble smaller
pieces of information, but we can never get all the pieces together.
As such, when information is lacking, we have to fall back on
inductive reasoning. This is another way of saying that we must
try to guess the odds.