A hosting provider reaches the rest of the internet in two ways. It buys transit from larger networks, which carry its traffic anywhere. It also arranges peering, where two networks exchange traffic directly, often at an internet exchange, usually without paying per megabyte.
More peering means shorter paths and lower delay to the networks where your visitors are. A provider with only one upstream is vulnerable if that carrier has trouble. This is hard to judge from a sales page, but a few speed tests from different places and a look at the provider's network map can tell you a lot.
Transit: paying for the whole internet
The internet is thousands of independent networks, each one an autonomous system with its own number. No small provider connects to all of them. Instead it pays a larger network to carry its traffic to any destination, and to announce the provider's addresses back to the world. That is transit, typically billed on the 95th percentile of traffic in a month: the provider samples usage every five minutes, discards the highest five per cent of samples, and pays for the rate at the next one down. The effect is that a short burst does not cost much, but a sustained rise does.
Transit gives complete reach, which is why every provider has it. Its weakness is that the traffic passes through someone else's network, on their terms, and you pay for each hop up the chain.
Peering: swapping traffic directly
If a hosting company and a large consumer ISP both send a lot of traffic to each other, it makes sense for them to connect directly rather than both paying a transit carrier to relay it. That is peering. The traffic between the two networks, and only that, goes over the direct link. Each side carries its own customers' packets to the meeting point, and normally neither charges the other.
Peering happens in two ways. A private interconnect is a cable between two networks in the same building, often used when traffic is large. A public exchange is a shared switch where dozens or hundreds of networks connect with one port each and agree peering sessions with whichever ones they like. Exchanges are usually run as non-profit or member-owned bodies and are found in most large cities.
What it means in numbers
Take a visitor in a city with a good exchange, and a hosting provider that is a member of it. Without peering, the request might cross the visitor's ISP, a national carrier, an international carrier and the provider's transit, perhaps eight or ten router hops and 35 ms (illustrative). With peering, the ISP hands the traffic to the provider at the exchange: three or four hops, perhaps 12 ms. Per request, 20 ms is invisible, but a page loading 60 objects over several connections feels the difference, and so does anything interactive.
The gains also show up in resilience. A provider with a dozen peering sessions and two transit providers can lose a transit carrier and lose very little, because much of its traffic did not use it anyway.
| Transit | Peering | |
|---|---|---|
| Reach | Whole internet | Only the peer's own customers |
| Cost | Per megabit, monthly | Port and cross-connect fees, no per-megabit charge |
| Path | Often longer | Usually shorter and more direct |
| Control | Carrier's policies | Agreed between two networks |
Verifying it
Run a trace to your provider from a few places and compare:
mtr -rwc 30 example.com
traceroute -A example.comThe -A option, where supported, shows the autonomous system number for each hop. Count the distinct numbers between you and the server: two or three is a short path, six or seven is a long one. Provider names in the hostnames often include an exchange's name, which suggests the connection is direct. Many networks also publish a peering policy and list the exchanges they attend; the information is public and a support team should be able to point to it. Then compare real response times from several regions, using a page-weight test such as the page weight tool together with your own measurements, and ask the support team directly.