Peering: The Handshake Nobody Signs

Somewhere in a data center, a fiber jumper connects two routers belonging to two different companies, and traffic starts moving across it in both directions. Nobody sends an invoice. That's the basic shape of settlement-free peering, and it has always struck me as one of the odder arrangements holding the internet together: two networks, often direct competitors, deciding it's cheaper and faster to hand traffic to each other than to keep paying a third party to carry it between them.

What surprises people outside the industry is how thin the paperwork usually is. Sometimes there's a real peering agreement, signed and filed, sitting in a folder nobody opens again. More often it's an email thread with a couple of AS numbers in it, and a BGP session that just starts working after a short phone call. I've seen peering arranged in less time than it takes to order coffee. When a contract does get written, it tends to arrive after the technical relationship already exists, catching up to a decision that was made somewhere else entirely.

That somewhere else is usually one person deciding whether to trust another person. A peering coordinator evaluating a request is really asking whether this network runs a clean shop and answers the phone during an incident. You can't find that answer in a legal document. It comes from having watched someone handle a bad night well, or badly, over several years.

Where the trust actually gets built

That's why the NOG circuit carries more weight than its line item in anyone's travel budget would suggest. I've spent a good chunk of my career in that world, and the pattern repeats at every event: the same faces, a little older each time, in the same style of hotel conference room with bad coffee and worse carpet. People stop introducing themselves by company after a while. The company might be different next year. The relationship is the part that persists.

The scheduled talks matter less than the program suggests. The real work happens in the hallway, or at the bar afterward, or in the five minutes before a session when someone leans over and asks if you've been seeing the same upstream flakiness they have. That conversation is doing something concrete. It's building exactly the kind of track record that gets cited later, informally, when someone decides whether to open a session with a network they've never technically done business with.

I'd go further and say the badge itself is doing engineering work, just work that's hard to distinguish from socializing unless you already know what you're looking at. An engineer who's been reliable and straightforward across a decade of these meetings has built something that functions almost like infrastructure. It shortens the distance between a problem and a fix. It's why a 2am call gets picked up by someone who already knows the account.

I don't want to oversell this. It's a clubby system, and clubby systems have an obvious failure mode: someone without a decade of hallway time is starting from a real disadvantage, whatever the quality of their network on paper. Trust economies work well for the people already inside them and less well for everyone else. Worth admitting rather than romanticizing.

But I don't think the informality is a bug the industry forgot to fix. A contract is good at recording what two parties agreed to on a given day. It's bad at telling you whether either party is still worth trusting six years later. For that you need people who've watched each other operate, in person, repeatedly. That's how a reputation gets earned instead of declared. The traffic crossing the interconnect is the visible part. Whatever made the interconnect possible happened earlier, off the record, in a room with bad coffee and a name badge.