Internet exchange points, where networks physically meet

The internet is metaphorically a cloud. Physically, it's wires and switches and routers in specific buildings, and a remarkably small number of those buildings carry most of the world's traffic.

Internet Exchange Points (IXPs) are the most important of these buildings. They are where independent networks plug in their routers to a shared switching fabric, peer with each other, and exchange traffic. Without IXPs, the internet would be slower, more expensive, and far less robust.

This article is about the physical reality of how the internet's networks actually meet.

What an IXP physically is

An IXP is, fundamentally, a piece of switching infrastructure operated by a neutral party in a specific physical location.

Picture: a large room (sometimes a series of rooms across a data center campus) with rack after rack of fiber-distribution panels, network switches, and routers belonging to various member networks. The IXP operator runs the switching fabric (often a few large carrier-grade switches that connect all members).

To join, a network:

1. Brings its router to a colocation suite inside the IXP's facility (or pays for "remote peering" via a third party that's already there).
2. Runs a fiber cross-connect to the IXP's switching fabric.
3. Pays the IXP a port fee (often $200 to $5000/month, depending on port speed: 10G to 400G).
4. Receives a VLAN on the shared fabric where it can peer with other members.

From the member's perspective, joining an IXP gives them one cross-connect that can reach hundreds of other members.

The peering fabric

The IXP's switching fabric is the centerpiece. Common configurations:

• Each member's router has at least one port on the fabric (their "peering port").
• The fabric is one large layer-2 segment (Ethernet), so members can directly exchange packets at line rate.
• IPv4 and IPv6 each get their own address space on the fabric.
• Members establish BGP peering sessions with other members over the fabric.

The IXP itself doesn't route traffic. It just provides the layer-2 fabric. All routing decisions are made by member routers using BGP.

Some IXPs offer additional services on top: a route server (which simplifies peering by letting one BGP session reach many peers), reflector services, statistics dashboards, etc.

The major IXPs

A few that dominate global traffic:

DE-CIX Frankfurt. Roughly 12 Tbps peak traffic. ~1100 networks. Frankfurt's data-center concentration makes it a critical European hub.

AMS-IX Amsterdam. ~10 Tbps peak. ~900 networks. Long-time European powerhouse.

LINX London. ~7 Tbps peak. Slightly fewer networks but with strong UK and trans-Atlantic links.

Equinix Internet Exchange. Network-of-networks across many cities (Equinix data centers worldwide host an EIX presence). Different city peers exchange separately.

DEC-IX New York. Growing rapidly. North American answer to European powerhouses.

HKIX Hong Kong. Major Asian hub.

JPNAP Tokyo. Major Asian hub.

Megaport. Cloud-friendly virtual IXP across many cities. Differs in model (everything is on-demand virtual ports).

Together, the top 20 IXPs handle 90+ percent of all peering traffic globally.

Why the geography matters

IXPs concentrate in specific cities for several reasons:

Submarine cable landings. Coastal cities where major cables come ashore (Marseille, Mombasa, Sydney, Tokyo) become natural IXP locations.

Population density. Many users in one area justifies infrastructure there.

Existing data centers. Where data centers already cluster, IXPs cluster too (Northern Virginia, Frankfurt, Singapore, Tokyo).

Regulatory environment. Some countries have onerous data laws or political instability that makes them unsuitable as IXP hubs.

The result: a small number of global hubs. Frankfurt, Amsterdam, London, Northern Virginia, Singapore, Hong Kong, Tokyo. Most internet traffic passes through one of these.

What being at an IXP buys you

For a network (hosting provider, ISP, content provider):

Lower transit costs. Traffic peered directly doesn't use your transit bandwidth. For a hosting provider with major peering, transit bills drop substantially.

Lower latency to peers. A peered destination is one short hop away. Transit might involve crossing the country or the ocean.

Better redundancy. If transit fails, peered paths still work. If peering fails (rare), transit fills in.

Operational visibility. You can see exactly which networks are at the IXP. Easier to negotiate with potential peers.

Reputation and trust. Being present at major IXPs is a signal that you're a serious network operator.

For a small operator, joining an IXP is a significant operational lift. Real router, real colocation, real BGP. But the payback can be quick if you have meaningful traffic to nearby peers.

Remote peering

Not every operator wants to own a router at every IXP. Remote peering lets you reach an IXP from another location.

Two main flavors:

Reseller / facilitator. A company (like Megaport, Console Connect) has a router at the IXP. They sell you a "virtual port" at that IXP. You connect to their network somewhere convenient (your own data center, your transit provider's facility), and they extend your reach to the IXP.

Tunneled peering. Some IXPs allow members to tunnel traffic in. Used carefully because of MTU and latency concerns.

Remote peering brings IXP benefits to operators who can't justify a full physical presence. Tradeoff: extra latency, less control, ongoing fees.

The route server

Bilateral peering means each pair of networks at the IXP sets up its own BGP session. With hundreds of members, that's hundreds of sessions per member.

A route server simplifies this:

• Each member peers with the route server (one or two sessions instead of many).
• The route server collects each member's route announcements and redistributes them to other members.
• Members get reachability to many peers via one BGP session.

Trade-offs:

• Easier to set up.
• Less granular control over which peer you accept from.
• Member's "true peer" is now the route server, not the originating member.

Most major IXPs operate route servers. Smaller operators use them; larger operators often peer bilaterally with their important partners for more control.

Route server policies

Networks can ask the route server to send them specific subsets of routes:

• All routes from all peers.
• Routes from specific peers.
• Routes with specific BGP communities.

This lets members fine-tune what they take, without each having to set up explicit bilateral peering with each desired source.

Public vs private peering

Two architectural patterns:

Public peering. Through the IXP's shared fabric. Cheap, easy, accessible. Lower bandwidth between any two members (you share the fabric).

Private peering. A direct cross-connect between two members, with dedicated bandwidth. More expensive (each side pays for their port and the cross-connect). Higher bandwidth and lower latency for the specific pair.

Most operators do both: public peering for many small peers, private peering for big partners (major content providers, major access networks).

What an IXP looks like in operation

A typical day at a large IXP:

• Members send terabits per second across the fabric.
• Most peering sessions hum along undisturbed.
• BGP updates flow when network changes happen.
• The IXP operator monitors switch health, port utilization, optical link quality.
• Members occasionally swap ports or upgrade speeds.

Incidents happen: a switch fails, a major link breaks. Major IXPs have redundant fabrics specifically to handle this. Failures are rare and well-handled.

When IXPs go offline

Rare but consequential. A major IXP outage instantly disconnects hundreds of peering sessions. Networks fall back to transit. Transit is more expensive and often slower. Users notice.

Major incidents:

• AMS-IX 2015. A configuration change caused a brief outage. Disruption across European traffic.
• DE-CIX 2018. A power failure took some segments offline. Recovery within hours.

These are rare enough that operators tolerate the residual risk. The dominant mode of IXP operation is reliable.

What IXPs mean for hosting

For a hosting company, IXP strategy is a real decision:

Which IXPs to join. Depends on customer geography. Hosting customers in Europe? Frankfurt and Amsterdam are obvious. Asia? Hong Kong, Singapore, Tokyo.

Public vs private peering with which partners. Major content providers (game studios, video, etc.) usually warrant private peering. Smaller peers via the route server.

How much port capacity. A 100G port at a major IXP is real money but lets a network grow without re-investment.

Hosting companies that take this seriously deliver visibly better network performance. Hosting companies that don't show it in their latency and uptime.

What IXPs mean for users

You don't interact with IXPs directly. But:

Your latency to specific services is shaped by peering paths through IXPs. If your ISP peers with a service at a nearby IXP, you have a short, fast path.

Your reliability is shaped by IXP architecture. Distributed IXP infrastructure means single-point failures are rare.

The cost of internet service is shaped by IXP economics. ISPs that peer well pay less for transit and can pass savings (or invest in capacity).

Coming up

We've covered IXPs as physical and operational entities. Next: a powerful technique built on top of all this peering and routing: anycast, where one IP is announced from many locations and BGP automatically routes users to the closest one.