Freya Sundby

I do the arithmetic in binary once and then convert, because the decimal form hides the boundary between the network part and the host part. The prefix length says how many leading bits belong to the network. So the network address clears every bit after the prefix and the broadcast address sets every one after it, and the usable range is what lies between. That is the whole of it, and every subnetting question reduces to counting bits. The two addresses to exclude catch people out. A prefix of thirty leaves two addresses and no usable hosts. A prefix of thirty-one leaves one, which is a valid entry and almost never what was wanted. A prefix of thirty-two is a host route. The classic question asks for the number of subnets and hosts, and the answer depends on which prefix length is in force, so the question has to state it. IPv6 is the same idea in a different shape. One hundred and twenty-eight bits written as eight groups of hexadecimal, with a rule that compresses a run of zero groups. The default prefix on a single interface is sixty-four, which halves the address, and the subnetting question has a different answer because an allocation is large enough for a site. The transition has a piece that did not change: the routing table and the subnet mask mean the same thing in either family, and the arithmetic is identical once the width is fixed. Every tool that takes both has to be told which one it is looking at. I also cover what to do when a mask and a prefix disagree, which is to fix the configuration rather than to reason about it.