In computer networking, peering is a voluntary interconnection of administratively separate Internet networks for the purpose of exchanging traffic between the customers of each network. The pure definition of peering is settlement-free or "sender keeps all," meaning that neither party pays the other for the exchanged traffic; instead, each derives revenue from its own customers.
Peering requires physical interconnection of the networks, an exchange of routing information through the Border Gateway Protocol (BGP) routing protocol and is often accompanied by peering agreements of varying formality, from "handshake" to thick contracts.
Marketing and commercial pressures have led to the word “peering” being also used routinely when there is some settlement involved, even though that does not correspond to the original technical meaning of the word. The phrase "settlement-free peering" is in turn used to unambiguously describe the pure cost-free peering situation.
The relationships between these networks are generally described by one of the following three categories:
Furthermore, in order for a network to reach any specific other network on the Internet, it must either:
The Internet is based on the principle of global reachability (sometimes called end-to-end reachability), which means that any Internet user can reach any other Internet user as though they were on the same network. Therefore, any Internet connected network must by definition either pay another network for transit, or peer with every other network who also does not purchase transit.
Peering involves two networks coming together to exchange traffic with each other freely, and for mutual benefit. This 'mutual benefit' is most often the motivation behind peering, which is often described solely by "reduced costs for transit services". Other less tangible motivations can include:
The physical interconnections used for peering are categorized into two types:
Public peering is accomplished across a Layer 2 access technology, generally called a shared fabric. At these locations, multiple carriers interconnect with one or more other carriers across a single physical port. Historically, public peering locations were known as network access points (NAPs); today they are most often called exchange points or Internet exchanges ("IXP" or "IX"). Many of the largest exchange points in the world can have hundreds of participants, and some span multiple buildings and colocation facilities across a city.
Since public peering allows networks interested in peering to interconnect with many other networks through a single port, it is often considered to offer "less capacity" than private peering, but to a larger number of networks. Many smaller networks, or networks who are just beginning to peer, find that public peering exchange points provide an excellent way to meet and interconnect with other networks who may be open to peering with them. Some larger networks utilize public peering as a way to aggregate a large number of "smaller peers", or as a location for conducting low-cost "trial peering" without the expense of provisioning private peering on a temporary basis, while other larger networks are not willing to participate at public exchanges at all.
A few exchange points, particularly in the United States, are operated by commercial carrier-neutral third parties. These operators typically go to great lengths to promote communication and encourage new peering, and will often arrange social events for these purposes.
Private peering is the direct interconnection between only two networks, across a Layer 1 or 2 medium that offers dedicated capacity that is not shared by any other parties. Early in the history of the Internet, many private peers occurred across 'telco' provisioned SONET circuits between individual carrier-owned facilities. Today, most private interconnections occur at carrier hotels or carrier neutral colocation facilities, where a direct crossconnect can be provisioned between participants within the same building, usually for a much lower cost than telco circuits.
Most of the traffic on the Internet, especially traffic between the largest networks, occurs via private peering. However, because of the resources required to provision each private peer, many networks are unwilling to provide private peering to "small" networks, or to "new" networks who have not yet proven that they will provide a mutual benefit.
Throughout the history of the Internet, there have been a spectrum of kinds of agreements between peers, ranging from handshake agreements to written contracts as required by one or more parties. Such agreements set forth the details of how traffic is to be exchanged, along with a list of expected activities which may be necessary to maintain the peering relationship, a list of activities which may be considered abusive and result in termination of the relationship, and details concerning how the relationship can be terminated. Detailed contracts of this type are typically used between the largest ISPs, and the ones operating in the most heavily-regulated economies. As of 2011, such contracts account for less than 0.5% of all peering agreements. (See examples below.)
The first Internet exchange point was the Commercial Internet Exchange (CIX), formed by Alternet/UUNET (now Verizon Business), PSI, and CERFNET to exchange traffic without regard for whether the traffic complied with the acceptable use policy (AUP) of the NSFNet or ANS' interconnection policy . The CIX infrastructure consisted of a single router, managed by PSI, and was initially located in Santa Clara, California. Paying CIX members were allowed to attach to the router directly or via leased lines. After some time, the router was also attached to the Pacific Bell SMDS cloud. The router was later moved to the Palo Alto Internet Exchange, or PAIX, which was developed and operated by Digital Equipment Corporation (DEC).
Another early exchange point was Metropolitan Area Ethernet, or MAE, in Tysons Corner, Virginia. When the United States government decided to de-fund the NSFNET backbone, Internet exchange points were needed to replace its function, and initial governmental funding was used to aid the MAE and bootstrap three other exchanges, which they dubbed NAPs, or "Network Access Points," in accordance with the terminology of the National Information Infrastructure document. All four are now defunct or no longer functioning as Internet exchange points:
As the Internet grew, and traffic levels increased, these NAPs became a network bottleneck. Most of the early NAPs utilized FDDI technology, which provided only 100 Mbit/s of capacity to each participant. Some of these exchanges upgraded to ATM technology, which provided OC-3 (155 Mbit/s) and OC-12 (622 Mbit/s) of capacity.
Other prospective exchange point operators moved directly into offering Ethernet technology, such as gigabit Ethernet (1000 Mbit/s), which quickly became the predominant choice for Internet exchange points due to the reduced cost and increased capacity offered. Today, almost all significant exchange points operate solely over Ethernet, and most of the largest exchange points offer ten gigabit Ethernet (10,000 Mbit/s) service.
During the dot-com boom, many exchange point and carrier neutral colocation providers had plans to build as many as 50 locations to promote carrier interconnection in the United States alone. Essentially all of these plans were abandoned following the dot-com bust, and today it is considered both economically and technically infeasible to support this level of interconnection among even the largest of networks.
By definition, peering is the voluntary and free exchange of traffic between two networks, for mutual benefit. If one or both networks believes that there is no longer a mutual benefit, they may decide to cease peering: this is known as depeering. Some of the reasons why one network may wish to depeer another include:
In some situations, networks who are being depeered have been known to attempt to fight to keep the peering by intentionally breaking the connectivity between the two networks when the peer is removed, either through a deliberate act or an act of omission. The goal is to force the depeering network to have so many customer complaints that they are willing to restore peering. Examples of this include forcing traffic via a path that does not have enough capacity to handle the load, or intentionally blocking alternate routes to or from the other network. Some very notable examples of these situations have included:
The "donut peering" model describes the intensive interconnection of small and medium-sized regional networks that make up much of the Internet. Traffic between these regional networks can be modeled as a toroid, with a core "donut hole" that is poorly interconnected to the networks around it.
As detailed above, some carriers attempted to form a cartel of self-described Tier 1 networks, nominally refusing to peer with any networks outside the oligopoly. Seeking to reduce transit costs, connections between regional networks bypass those "core" networks. Data takes a more direct path, reducing latency and packet loss. This also improves resiliency between consumers and content providers via multiple connections in many locations around the world, in particular during business disputes between the core transit providers. 
The modern Internet operates with significantly more peering locations than at any time in the past, resulting in improved performance and better routing for the majority of the traffic on the Internet. However, in the interests of reducing costs and improving efficiency, most networks have attempted to standardize on relatively few locations within these individual regions where they will be able to quickly and efficiently interconnect with their peering partners.
See also: List of Internet Exchange Points by size. The largest individual exchange points in the world are AMS-IX in Amsterdam, followed by DE-CIX in Frankfurt Germany and LINX in London. The next largest exchange point is generally considered to be JPNAP in Tokyo, Japan. The United States, with a historically larger focus on private peering and commercial public peering, has a much smaller amount of traffic on public peers compared to other regions which operate non-profit exchange points. The combined exchange points in multiple cities operated by Equinix are generally considered to be the largest and most important, followed by the PAIX facilities which are operated by Switch and Data. Other important but smaller exchange points include LIPEX and LONAP in London UK, NYIIX in New York, and NAP of the Americas in Miami, Florida.
URLs to some public traffic statistics of exchange points include:
A great deal of the complexity in the BGP routing protocol exists to aid the enforcement and fine-tuning of peering and transit agreements. BGP allows operators to define a policy that determines where traffic is routed. Three things commonly used to determine routing are local-preference, multi exit discriminators (MEDs) and AS-Path. Local-preference is used internally within a network to differentiate classes of networks. For example, a particular network will have a higher preference set on internal and customer advertisements. Settlement free peering is then configured to be preferred over paid IP transit.
Networks that speak BGP to each other can engage in multi exit discriminator exchange with each other, although most do not. When networks interconnect in several locations, MEDs can be used to reference that network's interior gateway protocol cost. This results in both networks sharing the burden of transporting each others traffic on their own network (or cold potato). Hot-potato or nearest-exit routing, which is typically the normal behavior on the Internet, is where traffic destined to another network is delivered to the closest interconnection point.
Internet interconnection is not regulated in the same way that public telephone network interconnection is regulated. Nevertheless, Internet interconnection has been the subject of several areas of federal policy. Perhaps the most dramatic example of this is the attempted MCI Worldcom/Sprint merger. In this case, the Department of Justice blocked the merger specifically because of the impact of the merger on the Internet backbone market (thereby requiring MCI to divest itself of its successful "internetMCI" business to gain approval.) In 2001, the Federal Communications Commission's advisory committee, the Network Reliability and Interoperability Council recommended that Internet backbones publish their peering policies, something that they had been hesitant to do beforehand. The FCC has also reviewed competition in the backbone market in its Section 706 proceedings which review whether advanced telecommunications are being provided to all Americans in a reasonable and timely manner.
Finally, Internet interconnection has become an issue in the international arena under something known as the International Charging Arrangements for Internet Services (ICAIS). In the ICAIS debate, countries underserved by Internet backbones have complained that it is unfair that they must pay the full cost of connecting to an Internet exchange point in a different country, frequently the United States. These advocates argue that Internet interconnection should work like international telephone interconnection, with each party paying half of the cost. Those who argue against ICAIS point out that much of the problem would be solved by building local exchange points. A significant amount of the traffic, it is argued, that is brought to the US and exchanged then leaves the US, using US exchange points as switching offices but not terminating in the US. In some worst-case scenarios, traffic from one side of a street is brought to all the way to Miami, exchanged, and then returned to another side of the street. Countries with liberalized telecommunications and open markets, where competition between backbone providers occurs, tend to oppose ICAIS.