Peering and Transit

Every packet crossing the Internet passes through networks owned by different organizations. At each boundary, someone made a business decision: pay for access, or trade it?

These decisions—multiplied across thousands of networks—create the Internet's structure. Two arrangements dominate: transit (paying for global reach) and peering (trading reach for free). Understanding them explains why your packets take the paths they do.

Transit: Paying for the Whole Internet

Transit is simple: you pay a larger network to carry your traffic anywhere on the Internet. Buy transit, and you can reach any destination—your provider handles the rest, either through their own connections or by passing traffic to their providers.

A regional ISP might purchase transit from a national carrier. The regional ISP announces its routes to the national carrier, who advertises them globally. Traffic destined for the regional ISP's customers flows back through the national carrier. The regional ISP pays for both directions.

This is the only product where you pay someone to take your stuff AND pay them to bring stuff back. That asymmetry is why peering exists.

How Transit Pricing Works

Transit pricing follows a few common models:

Committed bandwidth charges a fixed monthly fee for guaranteed capacity. Burst above it occasionally, fine. Sustain above it, pay overage.

95th percentile billing samples usage every five minutes, then charges based on the 95th percentile over the billing period. This tolerates spikes while billing for sustained load.

Flat-rate pricing provides unlimited usage up to port capacity for a fixed fee. Simpler, but often more expensive for variable traffic.

Prices vary wildly by location. In competitive markets with many transit providers, large customers might pay under a dollar per megabit per month. In remote locations with limited competition, prices can be 100 times higher.

Peering: Trading Reach for Free

Peering is an arrangement where two networks exchange traffic directly without payment. Each network can reach the other's customers without buying transit.

Settlement-free peering means neither party pays. Both invest in infrastructure to reach the peering point, but no money changes hands for traffic.

Peering works when both networks believe they receive roughly equal value. This doesn't require equal traffic—one network might send far more than it receives if the other values the content or destinations that traffic represents.

Public and Private Peering

Public peering happens at Internet Exchange Points (IXPs), where many networks connect to shared switching infrastructure. Establish a peering session across the IXP fabric without running dedicated circuits between facilities.

Private peering uses direct connections between two networks' facilities. Common for very high traffic volumes where IXP infrastructure would be insufficient, or when networks want guaranteed capacity under their direct control.

Many relationships combine both: public peering at shared IXPs, private peering where traffic volumes justify dedicated links.

Why Networks Don't Peer with Everyone

Networks maintain peering policies with requirements for potential peers:

Traffic minimums ensure the operational overhead of maintaining a peering relationship is justified by actual traffic exchanged.

Geographic presence might require peers to exist at multiple locations, ensuring the relationship provides broad value.

Network quality standards ensure peers maintain reliable networks that won't degrade service.

Ratio requirements might specify traffic should be reasonably balanced—typically within 2:1 or 3:1.

These protect networks from managing relationships that provide minimal value while ensuring basic quality.

The Economics

The decision to peer or buy transit is arithmetic:

Transit is operational expense. Traffic grows, costs grow proportionally.

Peering is capital expense—ports, cross-connects, routers. But once infrastructure exists, marginal cost per bit is essentially zero.

For networks with substantial traffic to exchange, peering is almost always cheaper. The crossover point occurs at surprisingly low traffic levels.

Beyond cost, peering often improves performance. Fewer hops, lower latency, more direct paths. The financial case usually dominates, but performance sweetens the deal.

Paid Peering: The Middle Ground

Between settlement-free peering and transit lies paid peering: networks exchange traffic directly, but one party pays.

This happens when networks want direct exchange (for performance or strategic reasons) but can't agree on settlement-free terms—usually because of traffic asymmetry.

The paying network typically pays less than equivalent transit rates. The receiving network earns less than full transit revenue. Both prefer it to the alternatives.

Paid peering is controversial because it blurs categories. Is it peering with a fee, or discounted transit? The distinction matters for routing policy and relationships with other networks.

When Peering Ends

Depeering—terminating a peering relationship—has consequences:

Traffic that exchanged directly must now route through transit providers. Costs rise for one or both parties. Performance may degrade.

Customers experience longer paths, potential congestion through alternative routes.

Depeering sometimes serves as negotiating tactic when networks disagree on terms. Notable depeering events have made news when major networks ended relationships, affecting Internet performance regionally or globally.

Internet Exchange Points

IXPs make peering practical at scale. Rather than running direct connections between every pair of networks, an IXP provides shared switching fabric.

Major IXPs host hundreds of participating networks. Connect once to the IXP, establish peering sessions with dozens or hundreds of other networks. The infrastructure investment for extensive peering drops dramatically.

IXPs charge for ports (and sometimes traffic), but costs are far lower than equivalent private interconnections. This makes broad peering accessible to smaller networks.

Route servers at IXPs aggregate routes from all participants and redistribute them, simplifying BGP configuration for networks peering with many others.

Route Selection: Following the Money

Networks configure BGP to prefer routes in this order:

1. Customer routes (networks that pay you for transit)—most preferred because they represent revenue
2. Peer routes (networks you peer with)—next because they're free
3. Provider routes (networks you pay for transit)—least preferred because using them costs money

This hierarchy isn't arbitrary. It's the entire economic logic of the Internet encoded in router configuration.

Hot potato routing hands traffic to peer or customer networks as quickly as possible, minimizing distance carried on your own network. Cold potato routing carries traffic longer before handing off—potentially better paths, but higher cost.

The Shift Toward Peering

The Internet has moved from transit-dominated toward more extensive peering:

IXPs have proliferated globally, making peering geographically accessible.

Transit prices have fallen, but peering remains cheaper for substantial traffic volumes.

Content providers have built extensive networks specifically to peer rather than purchase transit.

Even smaller networks increasingly peer where possible.

This hasn't eliminated transit—Tier 1 networks remain essential—but it's created a more distributed, mesh-like interconnection model.