Internet Service Providers peering policies fall roughly into four categories: Open, Selective, Restrictive, and no-peering. Those with an Open Peering Policy peer with anyone in any location(s), while Selective and Restrictive Peers use pre-requisites to screen prospective peers. One potential prerequisite is a “Not to Exceed” Peering Traffic Ratio which specifies the maximum allowable asymmetry in traffic exchange to be considered a “peer”. This of course excludes content heavy ISPs and Content Companies from settlement-free peering relationships. But does this make sense? Is there a technical and/or business rationale for using peering ratios in this manner?
Several hundred people attended the Peering Birds–of-a-Feather session at NANOG 35 where this issue was debated. During and after the debate, dozens of seasoned peering Coordinators discussed and debated the six strongest arguments for using peering ratios as a peering candidate discriminator. This paper highlights and discusses both sides of this heated debate.
(Editor’s note: The NANOG 35 Peering BOF Debate was held in the middle of the Level3-Cogent de-peering event that led to a partition across parts of the Internet. Since then, the Level3-Cogent peering has been re-established with no information released regarding the terms of that peering. However, an observant Peering Coordinator noticed that the Level3 Peering Requirements no longer include explicit peering ratios to be maintained .)
Obtaining Peering with companies with an Open Peering policy can be as easy as a phone call and typing a few lines of router configuration. The technical component of peering is pretty straightforward.
It is slightly more difficult to negotiate peering with a Selective Peer - one that has some peering prerequisites . During these initial peering discussions one is likely to hear a variety of peering requirements including traffic minimums, 24/7 NOC and contact information exchange, peering upgrade policies, number of interconnections and geographic distribution requirements, etc . There are rational technical and operational reasons for each of these. Occasionally, peering requirements include a “Not to Exceed” traffic ratio to ensure that aggregate peering traffic ratios will not exceed a particular ratio (quoted in Out:In – not to exceed 2:1 is a common requirement).
The Peering Coordinator Community put on a debate on the rationality of peering ratios as a peering discriminator at NANOG 35 in Los Angeles. During that debate, and during the subsequent informal debates afterwards, the consensus was that this metric was neither technically sound nor business rational .
For the duration of this paper we will present the strongest arguments on both sides of the issue. We present the strongest arguments for using peering traffic ratios as a peering discriminator, and then the counterpoint(s) discrediting the argument.
Sources and Notes ...
Ren Provo pointed this out to the author with reference to the new Level3 Peering Policy: http://www.level3.com/1511.html
See Appendix A for definitions of Peering, Transit, Peering Ratios, etc.
For some sample peering policies with Peering Ratios explicitly mentioned, see MCI (AS701) Peering Policy at http://global.mci.com/uunet/peering/ , AT&T (AS7018) http://www.att.com/peering/ , AOL ATDN (AS1668) http://www.atdn.net/settlement_free_int.shtml , Tiscali (AS3257) http://www.ip.tiscali.net/peering-policy.html or nac.net (AS8001) http://eng.nac.net/peering.html
The gist of these are specified in Appendix B. Rationale include: decrease time to repair by mutual maintenance and exchange of contact info, improved resiliency and headroom from multiple interconnect points and peering link upgrade policies, etc.
The debate audience vote was about 100 to 3 that the metric was not rational.
“Business Rational” is the author’s term to indicate the long-term rationality/irrationality of a practice from an overall business perspective.
There are two hidden assumptions within this claim: First, that there is no revenue realized as a result of hauling this content traffic, and second, that there is an increased load on the backbone by “hauling” this traffic over a peering (“for free”) relationship. Let’s consider these broader arguments with a diagram.
Figure 1 – Content Heavy ISP A Peers with Big ISP.
In this example (see figure 1 above), the Content Heavy ISP A peers with a Big ISP which has a large global network spanning the United States and Europe. In this example, the Content Heavy ISP A is peered with the Big ISP Network in a single location and has a 30:1 outbound to inbound traffic ratio . As is typical with today’s web (http) traffic, we see small requests from the Big ISP access heavy customers generate large responses from the Content Heavy ISP A.
Next (see figure 2 below) let’s consider an Access Heavy ISP A in the exact same situation: Access Heavy ISP A peers with Big ISP, and Access Heavy ISP A Customers send small requests across to Big ISP’s Content customers, which in turn send big responses back.
Figure 2 – After Access Heavy ISP A Peers with Big ISP
Notice that the Big ISP Backbone load is identical in these two scenarios.
Counter Argument #1 – Since the load on the backbone is the same in both scenarios, there is no technical rationale for discriminating against one versus the other – the load on the Big ISP is identical!
Counter-Argument #2: The Big ISP in these scenarios is hauling the traffic across the globe on behalf of its customers. Massive outbound asymmetry or massive inbound asymmetry is an artifact of the type of customer served, regardless of whether served via peering or transit relationships.
What is different between these two scenarios is who is paying and in which direction the traffic is loading the backbone. In the first scenario, the Big ISP Content Heavy customers are paying Big ISP to distribute its content to the entire Internet (including to Access Heavy ISP A customers). In the second scenario, the Big ISP Access Heavy customers are paying the Big ISP for Transit (access to the global Internet including the Content Heavy ISP A content). In both situations, Big ISP is being paid for the traffic exchange – it is not “hauling the traffic all over the world for free.”
Sources and Notes ...
30:1 isn’t out of the realm of possibilities. Richard Steenbergen (nLayer) points out "In a purely content to eyeball exchange, a typical outbound packet with payload is 1500 bytes, while the typical TCP ACK response is between 40-64 bytes." This is roughly a 30:1 ratio.
Counter Argument #1 - This is a valid observation, and perhaps a good argument for requiring a peer to spread the peering load across multiple geographically distributed interconnect regions.
This is not however an argument for using Peering traffic ratios to restrict Peering. As demonstrated earlier, Peering traffic ratios are irrelevant; eventually this traffic traverses your network to get to your customers. The asymmetry is merely a side affect of the type of traffic requested by or served on behalf of your customers.
One discussion with a large backbone provider highlighted the folly. This Peering Coordinator was proud that the peering ratio required a peer to “adjust” traffic to maintain an “appropriate” traffic ratio. How did the peer accomplish this shift? They redirected some portion of traffic to a transit provider who in turn forwarded that traffic directly back to the large backbone provider, onto the same router as the peer! There was no change in backbone load, simply a shift of traffic from one interface card to the other. The ISP mandating peering traffic ratios realized no cost savings, enhanced revenue, or performance benefits from this mandate. In fact, traffic now takes a more circuitous path to reach their customers yielding slightly worse performance.
Counter-Counter Argument – Multiple interconnect points may make the problem worse. With hot potato routing the large volume of content enters my backbone sooner! On the other end of the spectrum, “cold-potato” routing that some have advocated to distribute the load more evenly across the interconnect points simply does not scale. Cold Potato routing using Multi-Exit Discriminators (MEDs) requires breaking up aggregate routing blocks into more granular chunks that can be routed to a specific interconnect point. This breaks the scaling characteristics of CIDR, and can become an operational nightmare as load-balancing and fail-over scenarios are encountered. So, unless there is a way to effectively spread the traffic across the interconnect points without breaking things, I will have to carry the traffic further; more costs to me.
Counter-Counter-Counter Argument – But you will have to pick up that traffic somewhere, either via a transit link that costs you money or a peering link. I can get you that same traffic for free more directly so with lower latency. No matter what, you have to carry that traffic to your customers.
Counter-Argument #2 – The asymmetry is simply a short-term artifact of the currently popular HTTP web traffic. For the longer term, this will change. For example, there may be more peer-to-peer traffic, or other types of traffic that will swap the traffic asymmetry. Any policy based on traffic ratios is short-sighted.
Sources and Notes ...
Let’s understand this argument with another example.
Figure 3 – Before Eyeball Heavy ISP Y peers with Content Heavy ISP A
Pictured here (figure 3) we have a snapshot of a part of the Internet that includes a Content Heavy ISP A, his upstream ISP X, and a peer of ISP X that we call Access Heavy ISP Y. Being Content heavy means that the small requests coming in are answered with large responses. The directed lines and their width highlight the traffic flows between Access Heavy ISP Y customers who request content from Content Heavy ISP A. Assume that the upstream ISP X and ISP Y peer with each other, and that the Traffic Ratio between ISP X and ISP Y is 1:1 in this initial scenario.
Now let’s assume that Content Heavy ISP A peers with Access Heavy ISP Y (see figure 4 below).
The traffic between Access Heavy ISP Y and Content Heavy ISP A no longer traverses ISP X but instead traverses the direct interconnection. What is the impact? Migrating those large traffic flows away from the ISP X to the ISP Y interconnect make ISP Y appear to be more content heavy to ISP X than in the previous situation. Why? There is now relatively less content going to ISP Y from ISP X, while the same amount of content is coming from ISP Y. This makes ISP Y’s ratios with ISP X slightly worse, maybe 1.2:1 now, after the peering.
Conclusion 3A: If an Access Heavy ISP peers with a Content Heavy ISP, it can adversely affect its peering traffic ratios with its other peers.
Now let’s consider the opposite case – what happens when a content heavy ISP peers with an Access Heavy ISP? What happens to the Content Heavy ISP’s peering traffic ratios with its other peers?
In this next picture (figure 5) we have an Access Heavy ISP A that purchases transit from ISP X, who has a peering relationship with Content Heavy ISP Y.
Figure 5 – Before Content Heavy ISP Y peers with Access Heavy ISP A
Small requests from Access Heavy ISP A’s customers generate large responses from Content Heavy ISP Y’s content customers. Let’s again assume here that ISP X and ISP Y have a 1:1 peering traffic ratio to begin with.
Now ISP A and ISP Y peer with each other, resulting in the following:
Figure 6 – After Content Heavy ISP Y peers with Access Heavy ISP
All traffic between ISP A and ISP Y now traverses the direct interconnection, shifting a large chunk of traffic from the Y:X peering relationship to the Y:A peering relationship. Now ISP Y appears less Content Heavy (from ISP X’s perspective), improving his peering traffic ratios with ISP X and others perhaps to .8:1.
Conclusion 3B: If a Content Heavy ISP peers with an Access Heavy ISP, it can positively affect its peering traffic ratios with its other peers.
Counter-Argument: But herein we reveal the folly. (False) Conclusion 3: Peering traffic ratios are valid discriminators because they help maintain peering traffic ratios with other (existing) peers.
While demonstrable, THIS IS A CIRCULAR ARGUMENT for the general question at hand.
Consider the simplified restatement of this argument: “Peering traffic ratios are a valid metric because they support (other) peering traffic ratios.”
When confronted with this circularity the advocates offer argument #4.
Sources and Notes ...
This argument is business rational and comes in many forms:
Counter-Argument #1 - All peering and transit relationships are negotiable business relationships, but peering traffic ratios are probably not the right metric for determining the terms of the relationship.
Counter-Argument #2 - As discussed above, the cost of distribution is the same regardless of whether content or access heavy. A more appropriate negotiating discussion surrounds the incremental value provided to the content or access customers by the network in question.
Counter-Argument #3 - And remember, you will have to carry this traffic by some means (transit or peering) to get it to your customers. So when negotiating, you are just negotiating for a different delivery mechanism to your customers – this may not command a high price.
Sources and Notes ...
The author notes one comment a few years back from a Peering Coordinator “An email to peering @
Counter-Argument #1 - This comes up occasionally, and may appear to be a strong (albeit short term) argument for discriminating against peering with Content heavy ISPs. A company can essentially say we can only peer with Access Heavy networks due to backbone links in one direction being saturated. However, keep in mind what this signals to the Internet community: the congestion on the network is or will soon affect the customers of this ISP. Therefore, this ISP can not meet the needs of its customers and is probably delivering poor performance at peak periods. This may signal that you are neither a good peering nor a good transit provider candidate.
Counter-Argument #2 - This loading problem can best be solved with better traffic engineering, with the allocation of more resources to the backbone, or by temporarily denying peering until the needed upgrades are done. In any case, denying peering based on the peering traffic ratio rationale is the least attractive of all of these other options.
Sources and Notes ...
In this argument (seldom stated out loud) Peering traffic ratios are being used as an arbitrary barrier to peering.
Counter-Argument - One could just as easily specify higher traffic volume, more points of interconnect, a larger backbone capacity into more geographic regions including ones that don’t make any sense. This is not a strong defense for the validity of peering traffic ratios as a rational discriminator.
Sources and Notes ...
Peering Traffic Ratios as a Peering Discriminator present several problems that make them not business rational:
Sources and Notes ...
In the heated discussions surrounding this topic there were six flavors of arguments put forth to support peering traffic ratios as a peering selection criteria.
Argument #1 - “I don’t want to haul your content all over the world for free.” This argument is countered with the observation that, whether peered with a content heavy or an access heavy ISP, the aggregate load on the ISP backbone is about the same and that the ISP is being paid by its customers for providing access to that traffic. It is not hauling the traffic for free.
Argument #2 – “OK, but there is massive asymmetry here. Look at how many bits miles I have to carry your content, while you have only to deliver your content across the exchange point.” Regardless of whether peered with content or access the load on your network is about the same; the access or content traffic has to traverse your network to get to your customers. The asymmetry is simply a reflection of the currently popular HTTP traffic profile.
Argument #3 – “As an Access Heavy ISP, I don’t want to peer with Content Heavy ISPs, because doing so will screw up my peering traffic ratios with my other peers.” The analysis and diagrammed traffic flows in this paper prove this assertion true, however, the argument is circular: ‘Peering Traffic Ratios are valid criteria because they support Peering Traffic Ratios with others.’
Argument #4 – “I want revenue for carrying your packets.” While this argument is business rational, peering traffic ratios are a poor indicator of the value derived from the interconnection.
Argument #5 – “My backbone is heavily loaded in one direction – I don’t have $ to upgrade the congested part of my core without a corresponding increase in revenue.” This loading problem can better be solved with better traffic engineering, with more resources for the core or by temporarily postponing all additional peering until the core is upgraded. Introducing a peering traffic ratio requirement is at best a temporary solution and signals a poorly managed network – a bad image for peers and transit customers alike.
Argument #6 – “I don’t want to peer with anyone else. Peering ratios help me systematically keep people out.” This is an understandable (although seldom articulated out loud) argument but is a poor defense for peering traffic ratios per se since it is an arbitrary discriminator; one could just as well have specified the size of the backbone core, the market capitalization or debt structure, or an extremely large number of distributed interconnect points. Peering Traffic Ratios emerged years ago in the Internet as a peering candidate discriminator, but appear to have their roots in the PSTN world. Ratios reflect the telephony settlement model of buying and selling minutes, where the traffic difference between in and out is used for settling at the end of the month. However, the Peering Traffic Ratio discriminator model for the Internet interconnection fails because traffic ratios are a poor indicator of relative value derived from the interconnect.
This paper represents the collective set of arguments presented, debated, and discussed at NANOG 35 in Los Angeles. Particularly strong arguments were shared with the author either directly or indirectly from Peter Cohen (Telia Sonera), Richard Steenbergen (nLayer), Patrick Gilmore (Akamai), Vikas Mehta (AOL), Bryan Garrett (BellSouth), Brian Sutterfield (Cox), Kevin Mulvahill (Equinix formerly from @Home), Steve Wilcox (Telecomplete), Louie Lee (Equinix), Brokaw Price (Yahoo!), Joe Provo (RCN), John Curran, Hamish MacEwan (CityLink), Brandon Galbraith, Maureen Carroll (T-Systems), Vince Fuller (Cisco), Joshua Sahala (TWTC), Ren Provo (SBC) .
Sources and Notes ...
Richard Steenbergen articulated this argument at the NANOG 35 Peering BOF X debate.
Avi Freedman during the Peering BOF X Q&A following the debate.
A few folks challenged this point. One can peer traffic that leads to more congestion and greater loss, and this has occurred in the past. But, at least this is in the control of the two parties that will suffer as a result of poor performance, as opposed to using a third party to transport the traffic.
Internet Transit Pricing Historical and Projections
Index of other white papers on peering
Chief Strategy Officer, International Internet Exchange (IIX)
WIlliam B. Norton is a consultant, author of The Internet Peering Playbook: Connecting to the Core of the Internet, and highly sought after public speaker and international recognized expert on Internet Peering. He is currently employed as the Chief Strategy and VP of Business Development for IIX, a distributed exchange point provider. He also maintains his position as Executive Director and Consultant for DrPeering.net, a leading Internet Peering portal and consultancy. With over twenty years of experience, Mr. Norton focuses his attention on sharing his knowledge with the broader community in the form of presentations, Internet white papers, and private on-site consultations.
From 1998-2008, Mr. Norton’s title was Co-Founder and Chief Technical Liaison for Equinix, a leading Internet Peering Exchange and colocation center provider. From startup to IPO and until 2008 when the market cap for Equinix was $3.6B, Mr. Norton spent 90% of his time working closely with the Peering Coordinator Community. As an established thought leader, his focus was on building a critical mass of carriers, ISPs and Content Providers and documenting the core values that Internet Peering provided, including the Peering Break Even Point and Estimating the Value of an Internet Exchange.
To this end, he created the white paper process, identifying interesting and important Internet Peering operations topics, and documenting what he learned from the peering community. He published and presented his research white papers in over 100 international operations and research forums. These activities helped establish the relationships necessary to attract the set of Tier 1 ISPs, Tier 2 ISPs, Cable Companies, and Content Providers necessary for a healthy Internet Exchange Point ecosystem.
Mr. Norton developed the first business plan for the North American Network Operator's Group (NANOG), the Operations forum for the North American Internet. He was chair of NANOG from May 1995 to June 1998 and was elected to the first NANOG Steering Committee post-NANOG revolution.
Connect on-line: http://DrPeering.net – RSS: http://ask.DrPeering.net – e-mail: wbn@TheCoreOfTheInter.net
The Peering White Papers are based on conversations with hundreds of Peering Coordinators and have gone through a validation process involving walking through the papers with hundreds of Peering Coordinators in public and private sessions.
While the price points cited in these papers are volatile and therefore out-of-date almost immediately, the definitions, the concepts and the logic remains valid.
If you have questions or comments, or would like a walk through any of the paper, please feel free to send email to consultants at DrPeering dot net
Please provide us with feedback on this white paper. Did you find it helpful? Were there errors or suggestions? Please tell us what you think using the form below.
Contact us by calling +1.650-614-5135 or sending e-mail to info (at) DrPeering.net