Skip to main content

Analysis: Content Peering and the Internet Economy

Is content peering a mutually beneficial exchange between ISPs and content providers? Recent analysis found the long-term impact is negative for ISPs. Because content peering allows both ISPs and content providers to reduce expenses, it can be argued that it offers a mutually beneficial exchange. However, using an economic model of the Internet, a recent Alcatel-Lucent study clearly quantifies an economic imbalance between ISPs and large content providers, which results from expanded content peering. The net effect of widespread content peering is a smaller profit pool to be divided among ISPs, and ultimately, it creates a trend toward decreasing profits for all ISPs. This analysis of costs and revenues also indicates that ISPs have been forced to accept content peering as a means of coping with the rapidly increasing cost of transporting video traffic. It further suggests that the dominant traffic position of large content providers compels ISPs to accept this form of peering, with its unequal traffic exchanges.

Peering trends

The Internet operates with an implied “reciprocal agreement.” Each ISP opens its network to traffic from every other ISP, thereby losing some autonomy but simultaneously benefiting from the collective whole. In principal, this system works because each participant contributes something to the ecosystem and receives a proportionate benefit. ISP peering — Introduced as a way to reduce IP transit costs, peering was based on a similar reciprocal relationship between ISPs. It allowed peering partners to engage in symmetric traffic exchanges without fees. Because peering costs less than buying transit from a backbone provider, it initially exploded among Tier 2 and Tier 3 ISPs. They often had hundreds of peering partners who exchanged about the same amount of inter-domain traffic, so the relationship remained mutually beneficial. Content peering — With the phenomenal growth of video traffic, content providers proposed settlement-free peering with ISPs. Known as “content peering,” these new relationships are based on asymmetric exchanges where more content is sourced into ISP networks than is sourced by their networks, as shown in Figure 1. ISPs accepted content peering as a way to reduce their own transit fees for rapidly escalating volumes of video traffic. To understand the impact of rising costs, consider a scenario in which content from one ISP is requested by subscribers on a neighboring ISP’s network and vice versa. If video quality is 2 Mb/s Standard Definition (SD) — and has, on average, 100 concurrent viewers — the ISP’s transit cost is $20,000 per month, based on an average transit rate from five years ago of $100 per Mb/s per month. This makes the initial economic motivation for using content peering obvious for Tier 2 and Tier 3 ISPs.

Unanticipated consequences

As a result of the dramatic growth of content peering, transit expenses have fallen — from several hundred dollars per Mb/s to under $10 per Mb/s today. Clearly, this type of peering has provided temporary economic relief for ISPs, but the following long-term consequences were not anticipated:

  • Lost income — Content peering allows both content providers and ISPs to reduce transit expenses. However, it also causes ISPs to lose some of the income they once earned from selling transit to content providers (as shown in Figure 1), while content providers do not lose income. This is an important point in understanding ISPs’ discontent, as they see dramatic traffic growth without experiencing commensurate economic growth.
  • Loss of control — Exponential video growth has put content providers in a dominant traffic position. Google, alone, may represent 10 percent to 20 percent of transit traffic. These large content providers have also made significant investments in building national networks to reach their hundreds of peering partners around the world. With this strategy, they are buying less IP transit and reducing their costs, while generating more and more traffic. As a result, they have significant leverage over Tier 2 and Tier 3 ISPs that are struggling to better manage their expenses.

Before and after content peering — an economic comparison

To examine how peering is affecting the Internet economy, an economic model was used to compare two views, one before the rise of content peering and one after. The following sections outline the assumptions used to create formal equations for this analysis and then summarize the key findings. General assumptions This economic model considers the Internet to be a single network with one economic system, and its total revenue is the income from content providers and subscribers. The purpose of looking at the profit of the entire ecosystem is that individual ISPs may increase or decrease in profitability depending on their business model; however, if the total ecosystem profit is degenerating we can infer that more ISPs are growing than shrinking or vice versa. As it turns out, different classes of ISPs as well as their hierarchy (Tier) benefit differently from changes in Internet topologies or business models. Our key assumption is that the Internet is a “conservation system,” meaning no new revenues are generated: Transit fees are simply a way of sharing content and subscriber revenue — and are not truly a new source of income. Further, because one ISP’s transit revenue is another ISP’s transit expense, the net transit revenue for the ecosystem is zero and thus the revenue of this system is also its profit (refer to the Internet profit equation 1.1 below). Finally, the model assumes all traffic generated by content providers is equal to the traffic actually consumed by subscribers (which ignores control traffic, multicast and so forth). Internet economy before content peering In this view, content provider and subscriber revenues flow up the hierarchy of ISPs, as shown in Figure 2. Among the different classes of ISPs, economic incentives vary. Tier 1 (T1) bandwidth providers want to grow transit traffic, which is their primary source of income. Tier 3 (T3) access providers and content ISPs want to reduce transit traffic, which is their primary expense. Tier 2 (T2) providers want to minimize buying transit and maximize selling. Assumptions for this view — A portion of content provider revenue paid to a T3 ISP will be distributed upstream to a T2 ISP, as a transit expense for the T3 ISP. The amount will be a function of the volume of traffic on the link between the two ISPs. The transit provider will charge for the maximum of the up/down link traffic. Similarly, the revenue from the content provider, plus a portion of the transit revenue received by the T2 ISP, is passed up the hierarchy to the T1 ISP. The model assumes a “nominal” transit rate, measured in dollars per Mb/s charged by the T1 provider. The T2 ISP must charge more than the T1 ISP to generate a profit. And because the T3 ISP buys transit from the T2 provider, it must also add a profit margin on top of the rate it pays. Finally, the model assumes that subscribers pay a fixed monthly rate which does not vary with bandwidth consumption. Content providers generate equal amounts of traffic, and each ISP has an equal number of subscribers. The profit of each ISP based on the above topology and equation 1.1 is: Where ‘η’ is the transit rate (in $/Mb/s) billed by the T1 ISP and ‘t’ is the traffic generated by the content provider. Also ‘α’ is the margin charged by the T2 ISP and ‘β’ the margin charged by the T3 ISP. T2 ISPs resell transit bandwidth at a rate of (αη) in order for them to maintain a profit, and T3 ISPs resell transit they buy from the T2 at a rate of (β η) where ‘β’ must be greater than ‘α’ for the T3 to maintain a profit on transit. Applying the assumptions — In this sample case, the network illustrated in Figure 2 has one content provider for each T2 and T3 ISP. Each content provider generates equal amounts of traffic, and each of the four ISPs has 5000 subscribers and charges $30 per month per subscriber. Finally, the profit for transit resale for T3 and T2 providers is 10 percent. Then using equations 1.2 we can graph the profits for each ISP. The profits for this sample case are shown in Figure 3. Note that we use Tier notations (T1, T2, T3) instead of ISP. ‘T2’ reflects the sum of the profit of the two ISPs at that level. Similarly, ‘T3’ reflects the sum of the profit of the two T3 ISPs. The left side summarizes the split in profits for each ISP class: The T1 provider keeps 42 percent, followed by T2s (21 percent) and T3s (8 percent). The right side shows the profit of each ISP as a function of increasing content provider traffic. The slope of T1 profits grows at twice the rate of T2 provider profits and roughly five times the rate of T3 providers. Internet economy with content peering In this second view of the Internet economy, peering is added to the model’s equations as a cost component. To determine how much traffic content providers can “peer away,” peering candidates must be identified, based on which ISPs have an incentive to peer with them. In Figure 4, ISPs higher in the hierarchy than content provider B (CP B) receive revenue from CP B’s traffic, giving them no incentive to peer. However, ISPs on a different “branch” off the Internet backbone would pay transit for CP B’s traffic, which does create an incentive to peer with CP B. These likely candidates include ISP C and ISP A. (For content provider A ISP D and ISP B have an incentive to peer.) Assuming that ISPs A, B, C and D have the same number of subscribers, content provider B can peer away approximately half its traffic — 25 percent with ISP C and 25 percent with ISP A. As a result, CP B will buy only half as much transit from ISP B as it once did. This same approach can be applied to the remaining content providers to complete the possible peering relationships. These assumptions can be used to develop profit equations for the ISPs on each tier of the network. With content peering, the profit of an ISP will be reduced by two factors: The amount a content provider “peers away” with other ISPs and the cost of its peering relationship with other content providers. Of course, its profit can also be increased by a reduction in transit costs, assuming peering is cheaper than transit. To calculate the profit of an individual tier, the model assumes that content providers at T3 maximize their peering relationships. Based on the above we can derive a new set of profit equations that factor in the impact of content peering (see equations 1.3). Comparing these equations to (1.2) we note a new term Cp which is the cost of peering. We note that it is subtracted from the profit of T2 and T3 providers reducing their overall profit. (The detailed analysis of all calculations above can be made available upon request.)

Comparing the two views

When content peering is part of the equation, the ecosystem profits are 50 percent less than without content peering. In this hypothetical case the profit for the T1s literally goes to zero, while the profit for the T2/T3 ISPs varies but trends downward with increased content peering (see Figure 5). This analysis uses the same assumptions for transit rate, subscribers, and so forth that were applied to the earlier evaluation of the Internet economy without content peering. It also assumes that content provider traffic is fixed at 40 Gb/s per content provider, and the cost of peering will be one-fourth of the nominal transit rate. The graph below is based on equations (1.3). Figure 5 shows some interesting behavior in the individual profit curves as a function of increased peering. (The x-axis represents one content provider maximizing its peering followed by a second, third, and so on, starting with the Tier 3 content providers.) At any specific point in the evolution, a profit trend can go up or down, because one ISP’s gain is often another ISP’s loss. By looking at the total ecosystem profit, however, a downward trend is clear. The net result of “widespread” content peering appears to be a net loss for ISPs, as well as for the system as a whole. This conclusion can be generalized for any topology and any number of ISPs conditionally, based on the assumptions provided. Finally, profit is also driven by the number of subscribers. Thus, this analysis does not suggest that ISPs become unprofitable based solely on content peering. It only makes the case that profits decline with increased content peering. The goal of this article is to show that inequities exist not only between ISP and content providers but also between ISPs, as ISP profits vary according to their service model and Tier level. ISPs have for some time been moving away from being just bandwidth sellers to data hosts (or data center providers) and application sellers. Unless operators can recoup the lost revenues from content providers, it is likely that network operators will need to create usage caps. An optimal solution for network operators is to offer Internet differentiated services and expose network APIs that enable others to create value-added services. Content providers are limited today because of inconsistent Internet performance, and this will only get worse with increasing video content and the advent of cloud services. Differentiated Internet services would be key to enabling new cloud service paradigms and a better video quality of experience. To contact the author or request additional information, please send an email to


Bill Krogfoss

About Bill Krogfoss

Mr. Krogfoss led technical teams and is also an individual contributor in the areas of application performance modeling, video performance and QoE, Internet architectures and net neutrality, content delivery networks, IPTV, and metro transport. Mr. Krogfoss has published over 20 technical papers in BLTJ, IEEE, CNS, OFC and Wireless Congress. He has also filed 6 patents in the areas of content delivery and QoE. Bill has also been an invited speaker at BEREC (European Regulator Body ), Wireless Congress, IEEE conferences, and US Senate Antitrust Subcommittee on topics from net neutrality to video performance and application QoE. He holds a BSEE degree from St. Cloud State University, Minnesota.

Article tags