IP address classes

The original Internet Protocol specification from 1981 split the 32-bit address into a network part and a host part, and used the first bits of the address to say where the split was. A router could tell a network's size from the address alone, with no mask to carry around.

Two Class A networks were never handed out: network 0 is reserved, and 127 is the loopback block that gives every computer 127.0.0.1. The rest went, mostly in the 1980s and early 1990s, to the US government, universities and the large computer companies of the day, each receiving sixteen million addresses whether it needed them or not.

The three sizes had nothing in between. An organization with 300 computers was too big for a Class C (254 hosts) and got a Class B (65,534), wasting more than 99% of it. That mismatch, multiplied across thousands of organizations, is what exhausted the Class B space by the early 1990s and forced the change described below.

The private ranges that every home network uses today were defined in 1996 in classful terms, and the vocabulary stuck: 10.0.0.0/8 is a single Class A, 172.16.0.0/12 is sixteen consecutive Class Bs, and 192.168.0.0/16 is 256 consecutive Class Cs. That is why your router's 192.168.1.0/24 network is so often called "a Class C" even though no class has governed it for thirty years. Public vs private IP addresses covers what those ranges mean in practice.

The 1981 specification left the addresses beginning with 111 undefined, as an "escape to extended addressing mode". They were later split in two.

Class D addresses are not assigned to devices. They name groups: a packet sent to 224.0.0.251 is delivered to every device that has joined that group, which is how Bonjour and Chromecast discovery work on your Wi-Fi (that address is mDNS). Routing protocols use others in the 224.0.0.x range; IPTV systems use blocks higher up. You will see Class D addresses in a packet capture and in your router's IGMP settings, never on an interface.

Class E is the strange one: 268 million addresses, about one sixteenth of the whole space, that have never been usable. The reserved status is hard-coded into operating systems and routers, which refuse to accept a 240.x.x.x address, and proposals to reclaim the block have failed because fixing every device on the internet costs more than the addresses are worth. One address in it is special on its own: 255.255.255.255 is the limited broadcast address, meaning "every device on this local network", which is what a device uses to find a DHCP server before it has an address at all.

By 1992 two things were breaking at once. The Class B space was running out, because it was the only sensible size for most organizations. And the routing table in the internet's core was growing faster than routers' memory, because every Class C handed out to a small organization was a separate entry that every backbone router had to carry.

Classless Inter-Domain Routing, introduced in 1993, fixed both by making the network size explicit instead of implied. An address now travels with a prefix length, written 203.0.113.0/24 or as a subnet mask, and the prefix can be any length from /0 to /32. The organization with 300 computers gets a /23 (510 hosts), not a /16. A provider that holds a large block announces one route for the whole thing, and the core routers never see the individual customers inside it. CIDR notation explained covers how to read and calculate prefixes, and the subnet calculator does the arithmetic.

The change was invisible to most users because the old default masks are still valid prefixes: a former Class C is simply a /24 now. What disappeared is the rule that the first bits decide the size. Today a former Class A block is typically carved into thousands of smaller blocks owned by different networks, and neighboring former Class C ranges are aggregated into something larger. The IP lookup shows the actual prefix an address belongs to, which is the only size that means anything now.

Nothing on the public internet routes by class any more, but the concept survives in four places.

• Exams and textbooks. Networking certifications still ask for the ranges and default masks, partly as history and partly because the multicast and reserved ranges are still real.
• Software defaults. When you type a static IPv4 address into Windows' adapter settings, it fills in the subnet mask from the first octet: 255.0.0.0 for 10.x.x.x, 255.255.0.0 for 172.x.x.x, 255.255.255.0 for 192.x.x.x. The guess is often wrong for 10.x networks, which are usually subnetted, and typing the right mask is on you. Some routers and older routing protocols (RIP version 1) make the same assumption.
• Everyday vocabulary. "A Class C" is shorthand for a /24 among people who manage networks, and "a Class B" for a /16. Harmless, as long as nobody configures a mask from it.
• The ranges that never changed. Class D is still multicast and Class E is still unusable. If a device shows an address starting with 224 or higher, something is misconfigured.

For anything you configure yourself, use the prefix. Find your current address and mask with the local IP tool or your device's network settings, and treat the first octet as a hint about which private range you are in, nothing more.