Internet Routing Policies

Internet routing is not a giant GPS system trying to find the shortest geographical path. It is a policy system built on top of business relationships, trust assumptions, and autonomous decision-making between networks. Border Gateway Protocol, or BGP, is the mechanism that carries those decisions across the internet.

Networks choose routes based on policy

Every large network, or autonomous system, decides which routes it prefers. That preference might depend on local policy, customer routes, peering agreements, transit cost, path length, or operational safety. The result is that traffic often follows an economically sensible path before it follows a physically shortest one.

A provider usually prefers routes learned from customers because carrying customer traffic earns revenue. Peering routes may be preferred over paid transit because they reduce cost. Local preference often outweighs AS path length, which is why the route with fewer hops is not always selected.

What BGP attributes really do

Several attributes influence the chosen path. Local preference is a strong internal signal for outbound choice. AS path length can make one route look less attractive than another. MED can influence entry preference between neighbouring networks under certain conditions. Communities let operators tag routes with hints for policy handling, blackholing, or traffic engineering.

These are policy tools, not guarantees. Different networks interpret and prioritise them differently. That flexibility is part of why the internet can connect so many independent operators, and also part of why routing can look surprising from the outside.

Why routing incidents happen

Because BGP is based on announced reachability, mistakes and abuse matter. A route leak can accidentally advertise paths that were never meant to be transit routes. A hijack can announce prefixes it does not legitimately own, attracting traffic away from the real destination. Even when the intent is benign, the consequence can be outage, latency spikes, or traffic interception.

This is why operational controls such as prefix filtering, max-prefix limits, route validation, and RPKI matter. They reduce the trust surface, though they do not remove it completely.

Performance is entangled with policy

From an application perspective, internet performance depends partly on where your users are and partly on how networks exchange traffic. A cheaper transit decision can increase latency. Weak peering in one region can create congestion. Content delivery networks exist in part because routing policy and geography do not always produce the path your application would prefer.

The key idea is that internet routing is a negotiated policy landscape. Traffic goes where independent networks allow and prefer it to go. Understanding that explains why the internet remains resilient, and why it sometimes behaves in ways that look irrational if you assume the goal is simply the shortest route.