Peering is a voluntary interconnection between two independent networks, each an autonomous system, in which they exchange traffic destined for each other’s users and customers directly rather than through a third-party provider. Among internet fundamentals it is one of the two basic commercial relationships that hold the internet together, the other being IP transit, in which one network pays another for access to the entire internet.
Peering is usually settlement-free, meaning neither side pays the other because both benefit roughly equally. Its defining feature is scope: a peer announces only its own routes and those of its customers, not the rest of the internet, so peering never replaces transit on its own but reduces how much traffic has to use it.
How peering works technically
Two networks that agree to peer connect their routers, either over a shared switching fabric or a dedicated cable, and establish a Border Gateway Protocol (BGP) session. Each side announces the IP prefixes it is willing to reach through the link, and each side applies routing policies so that traffic to those prefixes prefers the peering path over paid transit.
Network operators commonly express this preference with local preference values in BGP: routes learned from customers rank highest, routes from peers come next and routes from transit providers rank lowest. This ordering follows the economics, because customer traffic earns revenue, peer traffic is free and transit traffic costs money.
| Relationship | Routes announced to the neighbor | Routes accepted from the neighbor |
|---|---|---|
| Customer to provider | Own and own customers’ routes | Full internet table or default route |
| Peer to peer | Own and own customers’ routes | Peer’s and peer’s customers’ routes |
| Provider to customer | Full internet table | Customer’s routes only |
Peers must also filter what they accept. Prefix filters, maximum-prefix limits and validation with RPKI route origin authorization protect against route leaks and hijacks, in which a misconfigured or malicious network announces address space it does not own.
Public and private peering
Peering takes place in two physical forms. Public peering happens at an internet exchange point (IXP), where many networks connect to a shared switching fabric and can peer with any participant over one port. Private peering uses a dedicated cross-connect, called a private network interconnect (PNI), between two networks, typically inside the same data center.
| Aspect | Public peering at an IXP | Private network interconnect |
|---|---|---|
| Physical link | Port on a shared switch | Dedicated fiber cross-connect |
| Number of peers per port | Dozens to hundreds | One |
| Typical port speeds | 1 to 400 Gbit/s | 10 to 400 Gbit/s, often bundled |
| Cost structure | IXP port and membership fees | Cross-connect and port costs |
| Best for | Many small and medium flows | Large, steady traffic between two networks |
Many exchanges operate route servers, which let a member establish a single BGP session and receive routes from all other participants that use the route server. This multilateral peering lowers the administrative effort compared with bilateral sessions negotiated one by one.
Peering compared with IP transit
Every network that is not a Tier 1 network needs transit to reach the whole internet, but it can lower transit volume by peering with the networks it exchanges the most traffic with. The two relationships therefore complement each other.
| Property | Peering | IP transit |
|---|---|---|
| Reachability | Peer and its customers only | The entire internet |
| Payment | Usually none, sometimes paid | Customer pays per capacity or usage |
| Commercial basis | Mutual benefit | Service contract |
| Latency | Often lower, fewer networks in path | Depends on provider routing |
| Operational effort | Many sessions and relationships | Few sessions with providers |
Tier 1 networks are defined by the fact that they reach the whole internet through peering alone, without buying transit from anyone. All other networks combine the two relationships, and large content and cloud providers now carry a significant share of their traffic over direct peering.
Peering policies and economics
Networks publish a peering policy that states who they will peer with and under what conditions. Contact details, policies and facility lists are commonly maintained in the shared PeeringDB database, which operators consult before requesting a session.
- Open policy: the network peers with almost anyone present at a shared location.
- Selective policy: the network sets requirements such as minimum traffic volume, presence in several locations or 24-hour operations.
- Restrictive policy: the network peers only in exceptional cases, typical of large incumbents that prefer to sell transit.
- Paid peering: one side pays for access to the other’s customers without receiving full transit.
Selective policies often include a traffic ratio requirement, for example that inbound and outbound volumes differ by no more than a stated factor. Content-heavy networks send far more than they receive, which has led to peering disputes in which congested links remained unupgraded and users experienced slow streaming until a commercial agreement was reached.
Benefits, risks and common misconceptions
Well-planned peering improves latency, resilience and cost at the same time, but it brings operational responsibilities that must not be underestimated.
- Peering is not free of cost: ports, cross-connects, colocation and staff time are real expenses.
- Peering does not provide full connectivity: transit or a full peering mesh is still required.
- Poor filtering spreads incidents: an unfiltered peer can leak routes and divert traffic worldwide.
- Capacity must be managed: a saturated peering link degrades quality for exactly the traffic it was meant to improve.
The common idea of the internet as a single network owned by someone is the biggest misconception that peering corrects. The internet is a web of tens of thousands of autonomous systems, and peering agreements are a large part of how those independent networks cooperate.
In plain words
Imagine two neighboring towns, each with its own road network. Instead of sending every car between them along a toll motorway owned by a third company, the towns build a direct road between their borders and agree that neither will charge the other for using it.
Peering is that direct road for internet traffic. Two networks connect to each other and swap traffic for free because both gain from it. Data takes a shorter path, arrives faster and costs less than sending it through a paid provider in the middle.
A real-world example
A regional broadband provider with about 200,000 subscribers notices that evening traffic peaks at roughly 80 Gbit/s, and analysis of its flow data shows that more than half comes from a few large video, gaming and cloud networks. All of it currently arrives through two paid transit providers.
The provider joins the nearest major internet exchange point with two 100 Gbit/s ports, connects to the route server and arranges bilateral sessions with the biggest content networks. For the largest one, which sends more than 20 Gbit/s at peak, it adds a private network interconnect in the same data center. Within a few months, most peak traffic arrives over peering, transit commitments can be reduced, and customers see lower latency to popular services during the busiest hours.