Explainer

What is an internet exchange point? A plain-English guide

Learn what an internet exchange point does, how peering works, and what to check before you connect your network to one.

By the Cloudverg engineering team · · 6 min read

What it is

An internet exchange point is a shared switch fabric inside a data center where networks connect to trade traffic directly. It is not an ISP and it does not sell transit. Think of it as a meeting room: the room does not carry your parcels, but it lets carriers hand them over without going through a central depot.

Members include ISPs, content delivery networks, cloud providers, mobile carriers and enterprises. Each member takes a port on the exchange, usually 1G, 10G, 40G or 100G, and a cross-connect to their own router. Some exchanges offer virtual ports through a partner, so you can peer without a physical presence. That helps smaller teams but adds a dependency.

Public peering happens on the shared fabric. You can peer with a route server, which is one BGP session that carries routes for many members, or with individual networks through bilateral sessions. Private peering is a direct cross-connect between two networks, often at the same facility, and it is not the same as exchange membership. Most networks use a mix: route server for reach, bilateral sessions for key peers.

An exchange is not a replacement for transit. Unless you peer with every network you need, you still need a transit provider or default route for the rest of the internet. The exchange gives you choices and shorter paths for the traffic you can peer.

How it works

Connecting starts with a port request and a cross-connect order. You need a public ASN, a provider-independent IP block and a peering policy. The exchange assigns IP addresses on the peering LAN for IPv4 and IPv6. You configure BGP on your router and bring up a session to the route server or to a peer.

The route server is the fast path. It runs BGP with many members and re-advertises their routes to you. It applies policy and prefix filters. You still need your own filters. A route server is convenient, but it is not a substitute for knowing who you peer with. Bilateral sessions give you more control: local preference, MED, communities and prefix lists per peer.

Routing decisions follow BGP rules. Local preference wins first, then AS path length, then origin, then MED. Hot potato routing is common: each network hands traffic off at the closest exit, so forward and return paths may differ. That is normal. It can surprise stateful firewalls and load balancers that expect symmetry. Check both directions with traceroute and a looking glass.

Security is a shared duty. Filter prefixes you accept and prefixes you announce. Use IRR objects and RPKI ROV where peers support it. Set max-prefix limits. Enable unicast RPF on your edge to reduce spoofed traffic. Many exchanges offer a blackhole community for DDoS and run a looking glass. Monitor session state and prefix counts with BMP or NetFlow.

Why it matters for a business

The business case is path control. A direct peering session can cut hops to a cloud region, a CDN or a SaaS provider. Fewer hops usually means lower round-trip time and less jitter. That matters for voice, video, real-time APIs and any workload where tail latency shows up in the user experience. You also choose the peers, so you can keep traffic on networks you trust.

Resilience is the second reason. If you have peering plus transit, a failure in one path does not have to take out the service. But peering is not automatically redundant. One port, one cross-connect or one facility is still a single point of failure. You need diverse ports, diverse routers and, for critical traffic, a second exchange or a private connection.

Cost is more nuanced than free peering. Peering itself is often settlement-free, but you pay for the port, the cross-connect, the colocation space, the router and the engineering time. The saving comes from carrying less traffic over paid transit. Run the numbers on your own traffic mix. If most of your traffic goes to one cloud, a private cloud connection may be simpler than an exchange port.

Visibility is the third reason. At an exchange you can see which networks you exchange traffic with, how much, and when paths change. It also helps in vendor talks: you can show which paths perform. The exchange does not fix your network, but it gives you more choices than a single transit default route.

Common mistakes

The first mistake is treating the exchange as transit. Peering gives you reachability to the peers you connect with, not to the whole internet. Keep a default route or a full table from a transit provider. Test reachability to a sample of destinations before you shift traffic.

The second mistake is skipping filters. Accepting all routes from a route server without prefix filters, RPKI checks and a max-prefix limit is an open door. A route leak or a hijack can send your traffic to the wrong place. Build an import policy that matches your peering policy. Build an export policy that only announces your own prefixes.

The third mistake is ignoring asymmetric routing. Hot potato routing means your outbound path may not match the return path. Firewalls, NAT and stateful load balancers can drop traffic when the return path changes. Test with traceroute from both sides. Use a looking glass at the exchange and at the peer. If you need symmetry, say so in the peering policy or use a private connection.

Other common slips: one port with no redundancy, no IPv6 peering when the exchange and peers support it, and no monitoring beyond a ping. MTU is another. The exchange fabric may support jumbo frames, but your path to the peer may not. Test with ping and the DF bit before you enable jumbo frames. Keep a runbook for session down and prefix count changes.

How to get started

Start with a goal and a measurement. Pick the destinations that matter: a cloud region, a CDN, a SaaS provider, a partner network. Measure current latency, jitter and loss with traceroute and a looking glass. If the paths are already short and stable, an exchange may not pay off. If you see tromboning through another city, it might.

Choose an exchange near your traffic or near your cloud region. Check the member list, route server support, port speeds, cross-connect process, IPv6 support, RPKI support, looking glass and PeeringDB records. Ask about the port contract and the cross-connect lead time. A virtual port through a partner can be a low-risk first step.

Prepare your side: ASN, IP space, IRR objects, RPKI ROAs and a peering policy. Order the port and cross-connect. Configure the route server session first, with max-prefix, prefix filters and RPKI ROV. Add bilateral peers one at a time. Test with looking glasses and traceroute from both sides. Monitor with BMP or NetFlow, and alert on session down and prefix count changes.

You can hire a network consultant, ask your transit provider for help, or use the Expert architecture session from Cloudverg at $450/month for a design review before you sign. The session includes a 90-minute call and a written summary. Whichever route you take, document the peering policy and review it each quarter. An exchange is a long-lived piece of your network, not a one-off project.

Frequently asked questions

Is an internet exchange point the same as an ISP?

No. An exchange is a shared switch fabric where networks meet. An ISP sells access and transit. You still need an ISP or transit provider for full internet reachability, unless you peer with every network you need.

Do I need my own ASN and IP addresses to connect to an exchange?

For public peering, usually yes. You need a public ASN and a provider-independent IP block to announce. Some exchanges or partners offer virtual ports, but you still need address space that peers will accept. Check the exchange requirements before you order.

What is the difference between a route server and bilateral peering?

A route server is one BGP session that gives you routes from many members, with the exchange applying policy. Bilateral peering is a direct session with one network, so you control the policy and traffic engineering. Most networks use both: route server for broad reach, bilateral sessions for key peers.

Monthly plan

Want help with this? Expert architecture session

$450/month

90 minutes a month · Within 5 business days

Ninety minutes a month with a senior architect on the cloud, AI or security question that is blocking you, plus a written summary.

Renews monthly. Cancel any time.
What is included

Related guides

More cloud networking guides   Cloudverg platforms

Smallest way to start

Get a senior architect on your problem this week

Ninety minutes a month on the cloud, AI or security question that is blocking you, a pre-read of your material and a written summary within two business days.

$450/month

Renews monthly. Cancel any time.