DHCP

A device does not wait for a lease to expire. It keeps two timers. At the first, it contacts the server that gave it the address and asks to extend. At the second, later one, it will accept an answer from any server on the network. The defaults are half the lease and seven eighths of it, so a device with a day-long lease asks again after twelve hours.

The standard sets no house value for the lease itself. An operator picks it, which is why the question "how long does an address last" has no general answer and needs measuring instead.

People assume a dynamic address is a restless one. The standard says otherwise. When a server picks an address it must try three things in order. First the address that device already holds, then the address it held before if that one is still free, and only then something new. A device that stays connected therefore keeps the same address through every renewal, and one that is switched off overnight usually gets its old address back.

Three things break that chain. The lease expires while the device is off and someone else takes the address. The operator renumbers on a schedule. Or the device moves to another network and is told its address is wrong. A carrier that puts many customers behind one public address is a separate matter, and so is a mobile network, where the address can move without anything being renewed.

We can measure the effect, because our free proxy pool remembers when each address was first seen and when it last answered. Of 794,459 addresses, 135,776 were alive at least once. Narrowing to consumer networks leaves 16,823, and their record is not what the folklore says. 59.7 percent were still answering more than 30 days after we first saw them. The average run was 56.3 days, and only 12.0 percent were gone within a day.

One correction on the way there, because it changed the answer. Our own datacenter flag had missed a hosting company, and its 4,683 addresses, which live 6.9 days on average, were sitting in the consumer pile. Dropping operator names that read as hosting moved the United States from 12.6 percent of addresses passing 30 days to 66.8 percent. A measurement is only as good as the population, and ours was wrong until we looked at the operators.

Where the line is matters more than anything else. Brazil, Indonesia and India keep an address for months. One North African incumbent hands out a new one almost daily: 86.7 percent of its addresses were gone within a day. That spread is operator policy, not protocol, and the protocol is the same everywhere.

The idea is older than the web. In September 1985 a protocol called BOOTP let a diskless machine ask the network who it was. The answer carried an address and a file to boot from. It had no notion of time: an address was assigned by hand, in a table, and stayed until someone edited it.

The lease arrived in October 1993, when the first version of this protocol added a period to the assignment. The version in use today was published in March 1997 and has not been replaced since. It is one of the oldest pieces of machinery still running under every home network. The address family is the only thing that moved on. IPv6 got its own version in November 2018. A device that finds no server at all falls back to a self-chosen address, a behaviour standardised in May 2005.

The word dynamic scares buyers into paying for static addresses they may not need. On the numbers above, a dynamic consumer address often outlives a rented server address, because renewal keeps returning the same one. A free proxy on such an address usually disappears because the machine stopped answering, not because its address moved.

What a static product buys is not a longer lease but the removal of the question. An address rented per period is held for you and does not depend on a device renewing anything. That matters when a target ties an account or a whitelist to the address. If it does not, a stable dynamic address is doing the same job quietly.