Version 1.0
March, 2001
Bill Woodcock
Packet Clearing House
 

This policy guide is intended to assist governmental business and technology development officers in the formation of national policy on Internet exchange facilities. It is intended primarily for countries which have either none, or no more than one preexisting exchange facility.

Background

The Internet is the "network of networks." It is not a centralized, organized system, but instead a mass of independently-operated businesses which carry data from one point to another by exchanging it at their borders. These borders can be simple and bilateral, or they can be rich meeting-points where many networks come to exchange data.

Most bilateral exchanges are private, between a smaller party and a larger one, and require a payment from the smaller party to the larger one in association with the exchange. Most exchanges which occur at public multilateral meeting-points, on the other hand, are between equals or "peers," and are "settlement-free," that is, they require no payment.

Public Internet exchange facilities, generally called "peering points" or "Internet exchanges," have allowed the growth of the commercial Internet as a business enterprise. Prior to the existence of Internet exchanges, there was a single "backbone" operated by the United States government, which interconnected all of the other networks which formed the Internet. The backbone placed restrictions upon commercial use, did not scale sufficiently to accommodate growth, and was retired in 1992.

Why host an Internet exchange within your country?

Internet service providers (ISPs) and Internet-dependent companies must either exchange traffic with their counterparts, or pay a larger ISP to perform that function on their behalf. This service is called "transit" and is one of an ISPís largest expenses. Most ISPs actually do both, exchanging as much traffic as possible settlement-free, and paying only for that portion which they canít exchange directly. By minimizing their transit, the portion which they have to pay another ISP to carry, they either retain profit or offer more competitive prices. Thus the most successful ISPs are those which exchange the greatest degree of traffic at peering points, and pay other ISPs as little as possible.

This is of particular interest to developing nations, since the larger ISPs which receive payments for carrying othersí traffic are all United States, British, and Japanese-owned. Thus countries which do not host healthy Internet exchanges domestically are subsidizing the growth of the Internet in more developed nations. This is, in effect, a regressive tax, and these payments are very substantial. Developing a strong domestic Internet exchange allows your domestic ISPs to retain capital and offer qualitatively better Internet access more affordably within your country.

The governmental goal of hosting an Internet exchange is two-fold: to reduce the export of capital to more developed countries through reverse-subsidy of telecommunications services, and to encourage the development of a competitive domestic Internet-enabled industry and the creation of high-paying knowledge-worker jobs which that entails. A secondary effect of a high degree of local traffic exchange is the formation of a strong content authoring and publication industry. That is, if most of what Internet users are viewing is coming from other domestic sources, itís likely to be domestically-created content, rather than advertisements from United States companies pushing United States products and cultural values.

This document describes a variety of strategies for encouraging the formation of a healthy Internet exchange and creating economic, regulatory, and tax strategies which support and strengthen it.

Specific Goals

An exchange will only succeed if it costs less than the alternative, which is paying other ISPs to carry traffic. Thus itís important to keep the cost of participating in an exchange low. There are many cost-components to exchange participation for an ISP. The first and most obvious is the fee charged by the exchange to the ISP, if any.

The cost of subsidizing an exchange point during its formation can be quite low, particularly relative to the eventual benefit. However an exchange point should not be artificially subsidized longer than need be, lest it discourage domestic competition from other parties which wish to operate their own exchange facilities at different price-points and with different features. Thus we recommend that governmental economic development agencies work with domestic ISPs to form a joint exchange-point governance body, and provide initial assistance, but withdraw from both organizational and financial involvement relatively quickly; certainly within eighteen months, unless very unusual circumstances obtain.

One goal should be the promotion of exchange points generally, rather than the subsidy of one specific government-run exchange. Exchange points generally do not need any special sort of tax incentives or exemption, since most incorporate as cooperatively-owned or self-owned non-profit entities, which are tax-exempt in most countries. One other necessary feature of the form of business entity which an exchange point takes is that it must have legal protection against acquisition by a for-profit business which could attempt to extract excess profit from the exchange, thus making it less competitive in its role as an exchange.

Other costs borne by ISPs who wish to participate at an exchange are that of the equipment which they need to put at the exchange, and the cost of the telecommunications lines or fiber between the exchange point and the ISPís principal place of business.

The ISPís cost in interconnecting their main facility ("POP" or point of presence) with the exchange is most likely to be a recurring cost in leasing telecom lines from the local PTT or a competitive carrier. This cost can be mitigated through subsidy of the exchange half-circuit (that is, through subsidy of the exchangeís end of the circuit, which has two endpoints, one at the exchange and one at the ISPís location). ISPs can reduce their costs more directly by simply locating critical portions of their equipment at the exchange point itself rather than elsewhere. ISPs can also work with competitive telecommunications providers to acquire long-term leases of dark fiber which they can operate by themselves at higher speeds over time, without an increase in costs to do so. Some combination of these three methods is probably appropriate.

Discouraging the Unnecessary use of Transit

A more controversial means of both encouraging ISPs to participate in an exchange and discouraging them from paying excessively for transit service is to manipulate the structure of the taxes levied upon the transit services to more specifically direct the behavior of both domestic ISPs and the foreign ISPs who offer transit. One could, for example, base the degree of taxation upon the difference between the price at which a foreign ISP offers service within your country and the price at which they offer the same service within their own parent country. For instance, if UUNet offers 1.5mbps of transit service for US$1500 in the United States, but for US$2000 in your country, they are attempting to extract a 33% surcharge from your domestic ISPs. If Sprint offers 1.5mbps of service for US$1000 in the United States, but for US$1200 in your country, thatís a 20% surcharge. A tax can be scaled accordingly, to discourage your domestic ISPs from buying services from providers which exact high surcharges, while simultaneously encouraging foreign ISPs to offer that transit service which your domestic ISPs do still need at a competitive price.

It is unfortunately unlikely that you will be able to convince any great number of foreign ISPs to participate in your exchange and exchange traffic with your domestic ISPs settlement-free, but any who do should be strongly encouraged and held up as examples. You may find more willing participants in the class of so-called "content peers," service providers which distribute information on cache servers which are dispersed as widely as possible, so as to be very near to Internet users, in the periphery of the Internet. Examples of this type of service provider are Akamai and Digital Island in the web-content market, and Nominum and UltraDNS in the domain-name service business. By bringing cached copies of content near to your countryís Internet users, they can greatly increase perceived Internet performance while decreasing your ISPsí need to pass that traffic through their transit providers. Content peers should also be strongly encouraged to participate in your exchange.

The Problem with Fracturing the Market

Commercial operators of exchange facilities (that is, ones who extract a profit from the operation of a facility, rather than operating it on a cost-recovery basis) are often under the misimpression that by refusing to interconnect the switch fabric at their location with those of their competitors, that they place their competitors at a disadvantage, and thus gain a relative advantage within the market. This is a misimpression, and actually increases costs and reduces value to all parties concerned.

Specifically, take as an example a region which has ten ISPs, each of which pay $1000/month to participate in an exchange switch-fabric, at one or more locations, and each of which contribute ten routes to the generally-available routing table. Each ISP is thus spending $1000 to gain access to ten routes from each of nine competitors, for a per-route price of $11.11. If a new exchange point operator decides to create a second, disconnected switch fabric, and charges $900 in order to draw ISPs away from the main switch fabric, he has fractured the market, and all ISPs either lose value, pay more, or both. Specifically, an ISP in this new environment has the choice of continuing to participate in the original exchange, moving to the new exchange, or participating in both exchanges. Letís say that of the ten ISPs, three choose to move to the new switch fabric and participate in it exclusively, three choose to participate in both switch fabrics, and four remain exclusively on the original switch fabric. The four which remain continue to pay $1000/month, but now have access to only 60 routes, for a per-route price of $16.67, plus the added cost of having to pay another ISP to carry their traffic to the three providers with whom theyíve lost contact. The three which moved to the new exchange pay only $900/month, but are only able to see 50 routes, for a per-route cost of $18/month, plus the cost of paying another ISP to carry their traffic back to the four providers theyíve lost contact with. The three which chose to participate in both switch fabrics donít have to pay another ISP to reach anyone, but their costs have gone up from $1000/month to $1900/month, or $21.11 per route. If a third switch fabric is added, costs go up even further.

There is a legitimate reason for wishing to run a disconnected switch fabric from an exchange point operatorís standpoint, which is that the integrity of the switch fabric can be compromised if ISPs which connect to it both misconfigure their equipment and disable the built-in safeties which protect adjacent devices. While a real concern, it can be protected against in other ways, and is scarcely sufficient to justify a doubling in usersí costs.

Since this is an unintuitive and often overlooked, but critical area of exchange economics, itís appropriate for governmental agencies to enforce regulation requiring that all exchange facilities within a metropolitan area interconnect with each other, forming a single contiguous switch fabric. Be watchful lest a fracturing occur due to unobvious causes, like exchange facility operators defining technical policies or requirements which discourage, limit, or constrain interconnection, or "colocation providers" which are not initially exchange operators, who subsequently install stand-alone switch fabrics.

At the same time, itís important to encourage exchange facility operators to compete and innovate. Diversity among the exchange facilities is an important quality; having a number of different operators providing interconnected exchange facilities at different combinations of price, performance, and reliability allows for a diverse market with a wide array of services available to consumers at competitive prices.

Note that exchange facility operators who charge primarily for switch fabric connections, more than for space, power, and cooling, are actually providing ISPs a disincentive to connect or to use their connection heavily, while failing to discourage them from wasting the primary constrained resources of the facility. Exchange operators should be encouraged to give away switch ports for free, regardless of the speed, while charging for space, power, and cooling. In this way, ISPs will be careful about congesting exchange pointsí physical facility, while exchanging traffic as freely as possible.

Lastly, exchange point operators should be strongly discouraged from adopting Mandatory Multi-Lateral Peering Agreements, or MMLPAs. An MMLPA requires that all exchange point participants exchange traffic with each other, and at no cost. While this is exactly the goal, making it an absolute prerequisite requirement of connection has a chilling effect, and tends to drive away all of the larger, most desirable, potential participants in the exchange. It also removes one of the strongest incentives to ISPs to maintain good technical practices. ISPs who fail to maintain good technical practices, or operate incompetently, endanger the stability of the exchange. In a normal exchange, without an MMLPA, such behavior is protected against by other ISPs isolating the misbehaving ISP until they correct their problem.

Business Environment Prerequisites

Since exchange points succeed or fail based almost entirely upon the degree of cooperation and investment by their participants, it is crucial to ensure that there are no obstacles to such cooperation and investment. One of the principal dangers is suspicion that investment on one ISPís part will be extracted to the benefit of a competitor. If ISPs believe this to be a possible outcome, they will protect themselves by not making an investment of time and energy in participating in an exchange in the first place.

The requirement of the government in this regard is that there must be a means of protecting the exchange business-entity against acquisition. That is, ISPs must be able to form a cooperative business structure for the operation of their exchange, which is guaranteed not to be acquired by another (presumably for-profit) company. An alternative is for the exchange to be operated by a business entity, often a university, which is not-for-profit (and can thus operate the exchange on a cost-recovery, rather than profit-extraction, basis), and is both not in the telecommunications sector and too large to be acquired by a company in that sector.

Another action governments can take to help the stability of exchanges is to offer a physical premises to house the exchange. Emergency-services call centers are often used, as are military facilities. This assists in assuring neutrality and that there can be no acquisition of jointly-created resources, and also assures participants of the long-term stability of the location. This allows them to amortize the costs of circuit or fiber installation over a longer period of time, and further decreases their cost of participation in the exchange. Top of page