Inter-ISP Bandwidth Markets[1]
Nemo Semret[2]
Abstract
The economic structure of Internet Service Provider inter-connections, a dichotomy between free peering and transit service models, is currently an inefficient bandwidth market. This inefficiency has a profound impact across the whole spectrum of the industry, from the profitability of service providers, to the evolution of new services and continued growth of the Internet as the network of networks. This paper synthesizes the key forces driving the wholesale (inter-ISP) market, the consequences of the inefficiencies, and how these forces can be practically harnessed to improve the economics of the industry for both suppliers and users of bandwidth.
Background
The global Internet is a collection of thousands of inter-connected networks. Each Internet Service Provider must connect to other ISPs to ensure that the traffic from its customers can reach its destination at any other ISP. This fundamental fact of the Internet landscape has two important dimensions: the physical topology and technological fabric of inter-connection on one hand, and the economic forces and business relationships that drive it on the other.
The physical interconnection between two ISPs is achieved in one of two ways:
· The two jointly lease a circuit directly between their respective points of presence; or
· Each establishes a connection to an inter-connection point[3] -- a facility operated by an individual ISP, a neutral third party, or as a cooperative entity between many ISPs.
The second approach has the advantage of scaling much better to large numbers of providers than the direct circuits approach. Indeed, it is more expensive for ISPs to connect one-to-one (for 10 ISPs, this requires 45 dedicated lines, one from each to every other), than to a common peering point (which requires 10 lines for 10 ISPs). In addition, other factors, such as access to content customers, and lower deployment and maintenance costs, have made the latter the preferred structure for inter-ISP connections.[6]
The business relationship in ISP inter-connections (regardless of the physical inter-connection mode) centers on the fact that accepting traffic from another provider implies additional bandwidth usage in one’s own network. This cost materializes either through congestion and lower quality for the traffic of the ISPs own customers, or through additional capacity investments to maintain quality. On the other hand, inter-connection between two networks allows each network’s customers to have access to the others’ customers, with better quality than if the traffic had to be routed through a third network. Thus, each party receiving traffic from another has to weigh the costs against the value of inter-connection. Traditionally, this has lead to ISPs adopting one of two models:
· Paid transit service, where one ISP (the seller) carries traffic from the other (the buyer) to any destination on the whole Internet. Aside for technical details, this service is essentially the same as the Internet access service offered by the seller to non-ISP customers, such as enterprises and content providers.[4]
· Free peering, where each ISP accepts traffic from the other destined to its own customers, including transit service buyers. The exchange is free of charge, based on the assumption that exchanged traffic is more or less equally valuable in both directions between the two ISPs.
Market Forces
Ideally, from the traffic flow point of view, all ISPs should be able to exchange traffic through peering with all ISPs that have a presence in the same region, and do so at exchange points where all are present. Indeed, peering allows data to travel to the destination using the best possible path – often directly to the destination ISP, whereas transit implies that traffic may have to travel through several ISPs and over longer and more congested paths than necessary to reach the final destination. This translates into lower quality for the end user.
Of course, since most networks do not have a presence in enough regions to sustain global connectivity with peering alone, buying transit service is essential for any ISP, except “Tier 1” providers such as UUNet, Cable & Wireless, Genuity, Sprint, AT&T, Level 3, and Global Crossing.
In today’s commercial Internet, the reality is that a non-Tier 1 ISP needs to connect to a Tier 1 network much more than vice-versa. Indeed, for the large ISP, the added value of being connected to one smaller network is marginal. But for the small network, peering with a Tier 1 backbone gives more direct access to a much larger set of destinations and sources of traffic. Because this economic reality is not reflected in the free peering model, in the late 1990s, Tier 1 providers started shying away from peering with smaller ISPs.[7]
“Over half our
business is in supporting the little guy, but there’s a resource drain every
time we peer with someone,” said John Sidgemore, UUNet’s president and chief
executive officer. “We just don’t want to do it for free.”[8]
Due to this lack of incentive, on the physical level, Tier 1 providers progressively abandoned “traditional” multi-lateral free peering points (known as “MAEs”), either by pulling out, or not upgrading capacity. On the economic level, the smaller ISPs increasingly have transit service as the only option for connecting to the Tier 1 ISPs, who continue to peer with each other at no charge, through private agreements. In recent months, particularly in Europe, hybrid models with paid partial peering have emerged[9].
The crude dichotomy which continues to prevail generally—either free peering for a restricted set of “equals”, or one way charging as a customer—has lead to a number of undesirable inter-actions between economics and traffic engineering. Generally, for lack of adequate pricing structures, Tier 1 ISPs use administrative measures, and rules-of-thumb to limit peering (see UUNet, Level 3 and Genuity peering qualification policies, published on their websites in early 2001). Almost all non Tier 1 ISPs seek to maximize peering relationships in order to reduce costs.
At the same time, transit service is believed to offer better quality; because the seller treats it is a revenue generating activity. The transit model suffers from lack of price transparency, and dynamics. Buyers and sellers negotiate based on imprecise perceptions of future traffic volume and value. The same party who, when dealing with an ISP whose customer mix is content oriented (and hence generates more outgoing traffic), claims that the traffic it receives is a load that the source should pay to have delivered; will, when negotiating with an ISP whose business is primarily Internet access, claim that the traffic it sends out itself is content that the receivers value and should pay to receive.
But, even though the trade-offs between transit and peering (and other hypothetical forms of IP inter-connection) lie at the heart of cost and revenues of all ISPs, in the day-to-day reality, the means to make an informed decision and implement the best inter-connection agreements are almost non-existent. In general, the pricing of an inter-connection (including the decision to peer free of charge) depends more on negotiating skill and experience than on the economic value of the traffic, and is often subject to non-disclosure agreements. Further, due to the complexity of, and uncertainty around, the negotiations, contracts tend to be longer term, and are difficult to renegotiate when the underlying conditions change.
Consequences and Perspectives
The inter-ISP bandwidth market is a Byzantine web of personal and corporate relationships, lacking two properties of efficient markets: price transparency and dynamics, and liquidity. The result of this inefficiency is that on the Internet today costs are higher than they need to be for ISPs and their customers. There are however deeper consequences, consequences that are shaping the evolution of the Internet.
In the absence of accurate price signals, capacity investments are not directed to where needs and value are greatest. Hence the following apparent contradiction: while backbone ISPs are said to have excess bandwidth (due to large expansion of fiber optic capacity in “core” long-haul networks), overall Internet traffic still experiences quality problems due to congestion (at inter-connection and access points).[10]
The peering choke points add a tremendous opportunity cost for ISPs, as evidenced by the existence of a new breed of service providers that have emerged over the last 3 years (Content Delivery Networks or CDNs, such as Akamai, Overlay Networks such as Internap, etc.). By replicating content at many locations accessible through multiple networks, these services allow the most valuable traffic to by-pass the inter-connection choke points (a fact often highlighted in their marketing literature). From the ISP perspective, this represents bandwidth intermediation between them and the buyers, and hence lost revenue, primarily due to the inefficiency of the inter-networking market. In 2001, CDNs charge on the order of 1 cent per MByte delivered[11]. Assuming “typical” burstiness (with a peak to mean ratio of 3), this is equivalent to about $1000/Mbps/month, while transit service sells for about $200/Mbps/month. Even factoring in delivery server software and other value adds, this represents a premium of several hundred percent. In other words, CDNs are skimming off the traffic for which customers have very high willingness to pay, precisely the high-margin traffic ISPs seek most in order to enhance their bottom line.
Differentiated grades of service cannot be rolled out, since they would require bandwidth to be allocated not just in one network, but on an inter-network basis for the customer to experience premium quality end-to-end. The technology for offering different types of service in IP networks has been developed and refined over the past decade, and, through standards such as MPLS and Diff-Serv, has been ready for deployment since the late 1990s. The business need exists as well, since offering higher-value services such as voice and video over IP is universally regarded as the key to ISPs improving margins. But since the demand for, and value of, higher quality Internet service classes is difficult to predict, adding the different service types to the inter-connection agreements is impossible with the current pricing structures.
The fundamental issue underlying all of the above is the lack of a sustainable economic model, one that maximizes the benefit of the Internet’s flexible architecture.
“The Internet
interconnection environment remains one where there are no soundly based models
of financial settlement in widespread use today. […] These issues […] will
shape the future of the entire global ISP industry.”[12]
The solution is an open, dynamic and efficient inter-ISP bandwidth market.[13]
All ISP inter-connections can be charged transparently, and according to the explicitly specified bandwidth in one direction or the other. Such a free market can cover the whole range of relationships dynamically rather than forcing all relationships into a few rigid categories (peering/transit, send-pay/receiver-pay).
Unlike other commodity markets, this market can be real-time, with delivery happening in seconds, thus allowing it to accurately reflect the changing patterns and economic value of traffic on the Internet.
The technical differences (in routing) between transit and peering need not be rigidly coupled with bandwidth pricing structures.
Bringing efficiency to the bandwidth market is critical to increasing the value provided by the Internet service provider industry today, and moving to real-time markets is the best way of achieving that.
Market size
Considering the value of the bandwidth transactions between ISPs (not including ISP-to-customer sales), we categorize and estimate the number of (US) ISPs and their inter-connections as follows, the objective being values for “order of magnitude” illustration purposes.
|
ISP
Category |
Tier 1 |
Tier 2 |
Tier 3 |
Total |
|
# of
ISPs |
10 |
1000 |
5000 |
|
|
Peering
points |
7 |
5 |
0 |
|
|
Peers per
point |
8 |
3 |
0 |
|
|
Peering
Connections |
56 |
15 |
0 |
|
|
Transit
Connections |
0 |
6 |
2 |
|
|
Traffic
per connection (Mbps) |
400 |
80 |
30 |
|
|
Traffic
per ISP (Mbps) |
22,400 |
1,680 |
60 |
|
|
Total
traffic (Mbps) |
224,000 |
1,680,000 |
300,000 |
2,204,000 |
|
|
|
|
|
|
|
Price
($/month/Mbps) |
200 |
200 |
200 |
|
|
Value
(M$/year) |
$ 538 |
$ 4,032 |
$ 720 |
$ 5,290 |
Traffic volume, which was doubling every 3-4 months in 1995-96, has since slowed to growing around 100% per year[14]. On the high end, the view is that as Internet penetration in business processes reaches critical mass, traffic growth has accelerated again, to doubling every 6 months.[15]
Internet bandwidth prices are of course opaque, but the consensus is that they are declining at a rate of 30 to 50% per year, with, unsurprisingly, the steepest declines being predicted by streaming software vendors and the lower declines by ISP equity analysts.[16] Based on the author’s experience with purchasing transit service, prices were around $300/Mbps/month in 2000 and in the low $200 in 2001.
For a baseline projection, we assume a price decline of 30% per year, and conservatively estimate traffic volume growth at 100% per year. This is consistent with independent estimates of the price elasticity of demand for bandwidth[17]. With these assumptions, the inter-ISP bandwidth market is estimated to be on the order of $5B/year in 2001 and in the $20B/year range in 2005.
|
Year |
2001 |
2002 |
2003 |
2004 |
2005 |
|
Total
traffic (Mbps) |
2204000 |
4408000 |
8816000 |
17632000 |
35264000 |
|
Price($/month/Mbps) |
200 |
140 |
98 |
68.6 |
48.02 |
|
Transaction
Value (M$/year) |
$
5,290 |
$
7,405 |
$
10,368 |
$
14,515 |
$
20,321 |