What is an Internet Exchange Point (IXP)?
An Internet Exchange Point (IXP) is a physical location where multiple autonomous systems connect to exchange traffic directly, without routing through third-party transit providers. IXPs are critical infrastructure that improve performance, reduce costs, and keep local traffic local — a packet from one city to another in the same country should not have to cross an ocean just because two networks never negotiated a direct connection.
Why IXPs Exist
Without IXPs, every network that wants to exchange traffic with another must either buy transit from a common upstream provider or negotiate a dedicated circuit between their facilities. For a small ISP in Frankfurt that wants to exchange traffic with a local CDN also located in Frankfurt, routing through a transit provider in Amsterdam and back is wasteful — it adds latency, cost, and unnecessary path length.
IXPs solve this by providing a neutral meeting point: a shared switching fabric in a carrier-neutral facility where any connected network can peer with any other connected network. The physical infrastructure — the switches and fiber in the data center — is operated by the IXP organization, which typically charges a port fee for connection but no per-traffic fee for the actual data exchanged.
How IXPs Work
An IXP operates one or more layer-2 switching fabrics — typically high-capacity Ethernet switches or switch stacks — housed in one or more carrier-neutral data centers. Member networks connect a router port to this fabric. Once physically connected, members establish BGP sessions over the shared layer-2 infrastructure.
There are two peering models at an IXP:
- Bilateral peering — Two members negotiate a direct BGP session between their routers on the IXP fabric. Each network explicitly chooses which peers to exchange routes with. This gives full control but requires negotiation and configuration overhead for each new peer relationship. A network at an IXP with 800 members would need up to 799 bilateral sessions to reach everyone.
- Route server peering — The IXP operates a route server (or a redundant pair), which is a BGP router that acts as a multi-lateral intermediary. Members peer with the route server instead of each other directly. The route server collects routes from all members and distributes them to all others (subject to filtering policies). A new member establishes one or two BGP sessions (to the route servers) and immediately has access to routes from all other route-server members.
The Peering LAN
All members of an IXP connect to a shared peering LAN — a single IP subnet (typically a /24 for IPv4 and a /64 for IPv6) that exists only on the IXP's switching fabric. Each member is assigned an IP address on this LAN for use as the BGP peering address. The peering LAN exists solely to enable BGP session establishment — actual traffic between members is exchanged over the same fabric but uses the members' own prefixes as source and destination.
Route Server Architecture
A route server (defined in RFC 7947) differs from a regular BGP router in one critical way: it does not modify the AS path of routes it receives. A normal BGP router prepends its own ASN when re-advertising a route; a route server passes routes through transparently. This preserves the original AS path, so a network receiving a route via the route server sees the actual originating AS and full path, not the route server's ASN in the middle.
Route servers implement per-member import and export policies. Members can configure which of their routes they want advertised to which other members, and can choose not to accept certain routes from specific members — all without disrupting other bilateral BGP relationships.
Major IXPs
IXPs range from small regional exchanges with a few dozen members to massive hubs handling multiple terabits per second:
| IXP | Location | Approx. members | Peak traffic |
|---|---|---|---|
| DE-CIX Frankfurt (AS6695) | Frankfurt, Germany | 1,100+ | ~14 Tbps |
| AMS-IX (AS1200) | Amsterdam, Netherlands | 900+ | ~10 Tbps |
| LINX (AS8714) | London, UK | 900+ | ~8 Tbps |
| Equinix IX | Global (50+ locations) | Varies by site | Multi-Tbps |
| NYIIX | New York, USA | 200+ | ~2 Tbps |
| JPNAP | Tokyo, Japan | 200+ | ~4 Tbps |
There are over 1,000 IXPs operating globally, including dozens of small regional exchanges in Africa, South America, and Southeast Asia that play an outsized role in keeping local traffic local and reducing dependence on expensive international transit links.
Economic Impact of Peering at IXPs
The economics of IXP peering are compelling. Transit bandwidth costs money — typically a per-megabit monthly fee paid to a transit provider. Every megabit of traffic exchanged directly at an IXP is traffic that does not flow through (and get billed by) a transit provider. For large content providers and CDNs that deliver enormous volumes of traffic, peering at IXPs saves tens of millions of dollars annually versus purchasing transit for everything.
From a latency perspective, traffic exchanged at a local IXP typically traverses a single Ethernet switch and the members' own border routers — a few hundred microseconds of added latency at most. The same traffic routed through a transit provider might add 5–50ms of round-trip latency depending on geography.
Impact on BGP Routing
When networks peer at an IXP, the AS paths between them collapse to two hops — your AS and their AS, with no transit ASes in between. This is directly visible in BGP data. Networks with extensive IXP peering — like Cloudflare (AS13335), which peers at over 300 IXPs globally — show very short AS paths from virtually anywhere on the internet.
Explore IXP-Connected Networks
You can observe the effect of IXP peering in the looking glass. Networks with extensive peering show short AS paths and many peer relationships: