🤝 BGP Peering Explained
How two Autonomous Systems agree to connect and exchange routes directly — settlement-free peering, paid peering, peering policies, and how to evaluate a peering session.
- Quick Answer
- Key Takeaways
- What Is BGP Peering?
- Why It Matters
- How It Works
- Architecture & Technical Detail
- Step-by-Step Process
- Visual Flow
- Practical Examples
- Real-World Use Cases
- Advantages
- Disadvantages & Risks
- Best Practices
- Security Considerations
- Performance Considerations
- Common Problems
- Troubleshooting
- Implementation Checklist
- Expert Tips
- Beginner Mistakes
- Comparison Tables
- Feature Table
- Key Terms Glossary
- FAQs
- Conclusion
Whether you're an ISP evaluating your first peering session, a content network trying to reduce transit costs, or simply curious how the internet's networks actually interconnect behind the scenes, peering is one of the most consequential — and most misunderstood — concepts in internet routing.
- Peering is a direct route-exchange agreement between two ASes, distinct from buying transit from an upstream provider.
- Most peering is settlement-free: neither side pays the other, since both benefit from the direct connection.
- Peering can be public (via a shared Internet Exchange Point) or private (a dedicated cross-connect between two networks).
- A peering session only exchanges routes for the peer's own network and customers — not full transit to the entire internet.
- Peering policies range from open (anyone can peer) to restrictive (selective, high traffic minimums required).
- Networks typically use a mix of transit and peering, shifting more traffic to peering over time as relationships grow.
🔍 What Is BGP Peering?
BGP peering is an agreement between two Autonomous Systems to interconnect their networks and exchange routing information directly via the Border Gateway Protocol, so that traffic destined for each other's networks (and typically each network's downstream customers) can flow between them without transiting a third party. Each side advertises its own routes — and the routes of any customers it provides transit to — but critically, does not advertise routes it learned from other peers or transit providers, since that would effectively make it a free transit provider for the peer.
This restriction is enforced through what's called a peer-to-peer export policy, sometimes summarized by the "no valley routing" principle in BGP route selection philosophy: traffic should only flow from customer to provider, provider to customer, or peer to peer — never peer to peer via a third network that would effectively be donating free transit. Understanding this constraint is fundamental to understanding what peering actually delivers, since it directly determines which destinations become reachable through a given peering session.
Peering exists as an alternative — usually a complement, not a full replacement — to buying IP transit from an upstream provider. Where transit gives you reachability to the entire internet routing table via a single paid relationship, peering gives you reachability to just one network's (and its customers') routes, typically at no cost, in exchange for reciprocal treatment. Most networks of meaningful size run a mix of both: transit for default, full-internet reachability, and peering to offload high-volume routes to specific networks directly.
The word "peer" implies rough parity between the two networks in question — comparable traffic volumes flowing in each direction — and this expectation of mutual benefit is precisely why most peering arrangements are settlement-free. Both networks save on transit costs for traffic that would otherwise need to flow through a paid upstream, and since the benefit is genuinely mutual, no money needs to change hands for the arrangement to make economic sense for both parties.
🎯 Why Peering Matters
Peering is one of the primary economic and technical forces that shapes how the internet's topology actually looks, and understanding it clarifies a lot about why certain networks are fast and cheap to reach while others aren't.
The most direct motivation is cost: every byte of traffic that flows over a peering session is a byte that doesn't need to be carried (and paid for) over a transit connection. For any network exchanging significant traffic with another specific network — a video streaming service and a large residential ISP, for example — establishing direct peering can save substantial money that would otherwise go toward transit bills, particularly as traffic volumes scale into the hundreds of gigabits per second seen at major exchange points.
Performance is the second major driver. A direct peering path is typically shorter (fewer hops), lower latency, and free of the congestion or additional cost pressure that can affect shared transit links. Content networks, cloud providers, and CDNs pursue aggressive peering strategies specifically because reducing the number of networks between their servers and end users directly improves the experience for those users — faster page loads, lower buffering, more responsive applications.
Peering also plays a structural role in internet resilience. A network with a rich mesh of peering relationships across many exchange points has more paths to reach the rest of the internet, meaning the failure or congestion of any single transit provider has a smaller overall impact. This diversity-through-peering is part of why major content and access networks invest heavily in peering infrastructure well beyond what pure cost savings alone would justify.
For smaller and mid-sized networks, peering is often the difference between being a purely "customer" network — entirely dependent on upstream transit providers for reachability and pricing — and having real leverage in the ecosystem. A network with an established peering presence at a major exchange point has more negotiating power with transit providers, since it has demonstrated it can reduce its transit dependency when the economics justify it.
⚙️ How Peering Actually Works
A peering relationship goes through a fairly consistent lifecycle from initial contact through an established, running BGP session.
Discovery and outreach
A network identifies a potential peer — often by observing significant traffic exchanged via transit, or by researching presence at a shared Internet Exchange Point — and reaches out, typically referencing the prospective peer's published peering policy.
Policy and traffic evaluation
Both sides evaluate whether the relationship meets their peering criteria: minimum traffic volume, geographic overlap, network size, and routing policy compatibility.
Interconnection method agreed
The networks agree whether to peer publicly (via a shared Internet Exchange Point's route server or direct VLAN) or privately (a dedicated point-to-point cross-connect, often within the same data center or carrier hotel).
Physical and logical setup
The physical cross-connect or exchange port is provisioned, IP addresses are assigned on the interconnect subnet, and BGP session parameters (ASNs, IPs, authentication) are exchanged.
BGP session established
Both routers configure an eBGP session to each other, applying appropriate import and export route filters, and the session comes up — routes begin exchanging automatically once the TCP-based BGP connection is active.
Validation and monitoring
Both sides confirm they're receiving the expected routes (and only the expected routes — not an accidental full table, which would indicate a filtering misconfiguration), and set up ongoing monitoring for session stability and traffic volume.
Once established, the session typically requires little ongoing maintenance — route filters may need periodic updates as either network's customer base changes, but the underlying relationship tends to be stable for years once set up correctly.
🏗️ Architecture & Technical Detail
Peering sessions run as standard external BGP (eBGP) sessions between routers belonging to the two different Autonomous Systems, typically directly connected at Layer 2 — either via a dedicated cross-connect cable or a shared VLAN at an Internet Exchange Point. The BGP session itself works identically to any other eBGP session technically; what differs is the route policy applied to it.
The defining technical characteristic of a peering session's export policy is that it advertises only "local preference" routes — the AS's own address space and the address space of any customers it provides transit to — and explicitly withholds routes learned from its own transit providers or other peers. This is typically implemented via BGP communities that tag routes by their origin (customer, peer, or transit-learned) so that export filters can be written declaratively rather than needing to be manually maintained as an exhaustive prefix list.
On the import side, a peering session's filters are similarly restrictive: a network should only accept routes for prefixes that the peer legitimately originates or provides transit for, rejecting anything that looks like the peer is unexpectedly announcing third-party address space — a defense against both misconfiguration and the kind of route leaks that have caused major internet outages in the past.
Two connection topologies dominate in practice. Private peering uses a dedicated physical cross-connect between the two networks' routers, typically within the same data center or carrier-neutral facility, offering guaranteed bandwidth and the highest performance for high-volume relationships. Public peering uses a shared switching fabric at an Internet Exchange Point, where a single physical port lets a network establish BGP sessions with many peers present on the same exchange, dramatically reducing the marginal cost of adding each additional peer.
🔧 Step-by-Step: Setting Up a Peering Session
Obtain your own ASN and address space
Peering requires your own Autonomous System Number and IP address space registered with a Regional Internet Registry — you can't peer using someone else's ASN.
Create a PeeringDB profile
Register your network on PeeringDB, listing your ASN, traffic levels, peering policy, and the exchange points and facilities where you're present — this is how most prospective peers will find and evaluate you.
Identify and reach out to prospective peers
Use traffic analysis, PeeringDB, and exchange point participant lists to identify networks worth peering with, then contact them via their published peering email or process.
Agree on interconnection method
Decide between public peering (via a shared exchange) or private peering (a dedicated cross-connect), based on expected traffic volume and shared facility presence.
Exchange session details
Share ASNs, interconnect IP addresses, and any authentication requirements (like MD5 or RPKI expectations) needed to configure the BGP session on both sides.
Configure filters before enabling the session
Build explicit import and export filters (ideally based on the peer's published IRR route objects) before bringing the session up, rather than accepting everything by default.
Bring up the session and validate
Enable the BGP session, confirm the expected number and content of routes are received, and monitor traffic to confirm the peering is functioning as intended.
✅ Implementation Checklist
Use this checklist when planning, negotiating, or auditing a peering relationship, whether it's your first session or your fiftieth.
- Own ASN and address space registered — confirmed with your Regional Internet Registry before any outreach begins.
- PeeringDB profile created and current — traffic levels, policy, and facility presence all accurately listed.
- Candidate peers identified from actual traffic data — not guesswork, but real flow analysis showing meaningful volume.
- Prospective peer's policy reviewed — minimum traffic requirements and preferred exchange points checked before contact.
- Interconnection method decided — public (exchange) for lower volume, private (cross-connect) for high volume.
- IRR route objects registered — so the peer (and everyone else) can build accurate import filters for your address space.
- Export filters built from communities — not manually maintained prefix lists that will drift out of date.
- Import filters scoped to the peer's legitimate space — rejecting anything outside their registered footprint.
- Maximum-prefix limit configured — a safety net against an unexpectedly large announcement from the new session.
- RPKI validation enabled where possible — an additional layer of protection against invalid origin announcements.
- Session monitoring in place — alerting on session state changes and unexpected route count shifts.
- Contact details documented — both technical and NOC contacts recorded for future troubleshooting.
📚 Key Terms Glossary
- Autonomous System (AS)
- A network or group of networks under a single administrative entity, identified globally by a unique Autonomous System Number and operating a unified routing policy toward the rest of the internet.
- Settlement-free interconnection
- An interconnection agreement, whether public or private, where neither party pays the other, based on the mutual benefit both networks receive from the direct connection.
- Cross-connect
- A dedicated physical cable, typically within a shared data center or carrier-neutral facility, directly connecting two networks' equipment for private peering or other point-to-point interconnection.
- Route server
- Infrastructure operated by an Internet Exchange Point that lets a member establish one BGP session to exchange routes with many other participants at once, simplifying multilateral peering.
- Export policy
- The rules a network applies to determine which routes it advertises to a given BGP neighbor, central to correctly scoping a peering session to only the peer's own network.
- Local preference
- A BGP attribute used to rank multiple candidate paths to the same destination, commonly configured so peer and customer routes are preferred over transit-learned routes.
- PeeringDB
- A free, community-maintained database where networks publish their ASN, traffic levels, peering policy, and facility or exchange presence, serving as the primary discovery tool for prospective peers.
- Peering policy
- A network's published criteria — traffic minimums, geographic requirements, contact process — for evaluating whether to establish a peering relationship with another network.