Names to addresses

Think of it when a site works for some people and not others after a move, or a new web address does not open yet.

In everyday life

A friend moves house and updates the post office. Neighbours still have the old address in their address books, and their letters go there until each of them updates the book.

One case

A school moves its website to a new server at 09:00 and points its name at the new address. Computers around the world saved the old address for up to 4 hours, the record’s time to liveHow long a saved copy is trusted before it is fetched again. Engineers write it TTL.. At 11:00 about half of all visitors still go to the old server, which is already switched off. By 13:00 every saved copy has run out.

Visitors moved to the new server over 4 hours, not at once
The case, cell by cell1 cell = 1 hour
Visitors meant to reach the new server (%)
100100100100100
Visitors who reach it (%)
0255075100
Hours of split traffic
4
On the old server at 11:00
50%
The share grows as each saved copy of the old address runs out.

Every case

A request to a name starts with a lookup, and the answer is saved at many layers, for example the phone, the home router and the internet provider. A change to the name spreads only as those copies run out. So a change is slow and uneven, and it is slow to undo too: the old answer lives on in places you cannot clear.

The word

Engineers call this

DNSThe internet’s phone book. It turns a name people can remember, such as school.example.org, into the number address computers use to find each other.

Lower the time to live before a move, at least as far ahead as the old time to live, so saved copies expire fast. Keep the old server running until every old copy has run out.

Where this idea leads