Your packets do not take the route you think they do

There is a short answer about internet routing and a useful one, and they are not the same. What follows is the useful one.

Networks negotiate, then route

The internet is tens of thousands of independent networks that exchange routing information through the border gateway protocol. Which path a packet takes reflects peering and transit agreements between those networks, which are commercial arrangements. This is why traffic between two cities in the same country sometimes travels through another country entirely.

It is not inefficiency so much as the visible result of who has agreed to carry whose traffic.

You are usually talking to a cache

Content delivery networks place servers inside or adjacent to consumer networks and serve copies of popular content from there. The effect is that a request for a large site typically terminates a few milliseconds away rather than at a distant origin.

Mechanically, this is why a site hosted on another continent can feel local, and why it stops feeling local for anything genuinely dynamic. Cache misses are the visible edge of the arrangement: the first request for an unpopular object travels all the way to the origin, which is how one obscure page on an otherwise fast site turns out to be slow.

Latency is mostly physics and hops

Light in fibre travels at roughly two-thirds the speed of light in vacuum, which sets a hard floor on round-trip time by distance. Each router in the path adds processing delay, and congested links add queuing delay on top.

In practice, increasing bandwidth does nothing for latency, which is why a faster connection does not make a distant server respond sooner. The first stretch is often the largest single contributor, because access technologies schedule transmissions in fixed intervals and a packet waits for its turn before it reaches the fibre at all.

Traceroute is informative and misleading

Traceroute shows one path at one moment, and routers may deprioritise the responses it depends on, producing alarming numbers at intermediate hops. Asymmetric routing means the return path can differ entirely from the outbound one, and traceroute only shows one direction. A high latency at hop five with normal latency at hop ten means the hop-five router was busy answering, not that the path is broken.

Traffic is also spread across parallel links, so consecutive probes can follow different paths and the printed route may be one that no single packet ever took.

Where it goes wrong

Route leaks and hijacks, where a network advertises addresses it does not own, have repeatedly redirected large volumes of traffic. Mitigations exist, notably route origin validation, and adoption is incomplete.

In practice, the protocol was designed for a network where participants trusted each other, and that assumption has not held. Undersea cable damage is the other routine failure, and because repair ships take days to weeks the symptom is usually a sudden reroute onto a longer path rather than an outage anyone announces.

Implementations differ, and vendors are not obliged to document the differences.

One address can be in many places at once

Anycast lets the same address be announced from dozens of sites, and routing simply hands each packet to whichever announcement looks nearest by its own commercial logic. This is how public resolvers and large platforms answer from a machine near you without you ever choosing one.

It also means a service can be perfectly healthy for you and unreachable for someone in the next city, because you are not talking to the same machine and no shared status page reflects that. Geolocation databases mislabel these addresses routinely, which is why a site occasionally decides you are in a country you have never visited and serves the wrong language or refuses entirely.

The takeaway

The route is a business decision. The latency is physics.

Understanding the failure mode tells you more than the feature list does.