Classful Addressing and Its Limits

The Five Address Classes

An IPv4 address is 32 bits, written as four decimal octets like 192.168.1.10. The classful model split those bits into a network portion and a host portion, and the location of that split was decided entirely by the leading bits of the first octet. A router could therefore classify any address just by inspecting its high-order bits.

Figure 3.1: Leading-bit decision tree for IPv4 classes A–E

Classes A, B, and C were the assignable classes. Each carves the 32 bits differently, and each "minus 2" in the host count reflects the network address (all-zeros host) and the directed broadcast address (all-ones host), neither of which can be given to a real device.

Figure 3.2: Network/host split of the 32-bit address per assignable class

Worked example. Take 172.16.40.5. The first octet, 172, falls between 128 and 191, so it is Class B, default mask 255.255.0.0 (/16). The first 16 bits (172.16) are the network and the last 16 (40.5) are the host. For 200.1.2.3, the first octet 200 lands in 192–223, so it is Class C, default mask 255.255.255.0 (/24), network 200.1.2.0, host 3.

Class D and Class E are never assigned to hosts and are not split into network/host portions. Class D (224.0.0.0/4) is reserved for multicast: a single stream sent to a group address reaches every host that joined the group, without the sender copying it per recipient. Think of unicast as a phone call, broadcast as shouting in a room, and multicast as a subscription newsletter written once. Class E (240.0.0.0/4) is reserved for research and experimental use, with 255.255.255.255 carved out separately as the limited broadcast address.

Special and Reserved Ranges

Not every address in the 32-bit space is up for grabs. Certain blocks are permanently reserved for special jobs, and you will meet them constantly in real work.

The most important set is RFC 1918 (1996), which reserves three private blocks for use inside private networks. Because they never appear on the public Internet, any organization can reuse them freely — which is why your home router hands out 192.168.x.x addresses. A device with an RFC 1918 address must use NAT to reach the public Internet. Think of these like internal room numbers: thousands of buildings all have a "Room 101," and that is fine because the number only means something inside the building.

The entire block 127.0.0.0/8 is reserved for loopback: any packet sent there loops back to the sending host and never leaves the machine. 127.0.0.1 is universally "localhost." This reservation is also why assignable Class A stops at 126 — the 127 block was already claimed. Meanwhile 169.254.0.0/16 is APIPA / link-local: a host set for DHCP that can find no server auto-assigns itself an address here so it can at least talk on the local link. Seeing a 169.254.x.x address is a strong hint that DHCP failed.

Finally, the documentation ranges (RFC 5737) — 192.0.2.0/24, 198.51.100.0/24, 203.0.113.0/24 — let manuals use realistic addresses that will never collide with real space, and 100.64.0.0/10 (RFC 6598) gives ISPs shared space for Carrier-Grade NAT without colliding with customers' RFC 1918 addresses.

Why Classful Failed

The classful model was easy for routers to parse, but it had a fatal flaw: only three coarse, fixed sizes. Class C held 254 hosts (too small for most), Class B held 65,534 (usually far too large), and Class A held 16.7 million (astronomical). There was nothing in between. An organization needing 2,000 hosts could not use a Class C, so it took an entire Class B and wasted roughly 63,000 addresses — locked away from everyone else. It is like a shoe store stocking only child size 1, adult size 9, and clown size 30: a size-11 foot must take the size 30.

This granularity problem cascaded. Because Class B was the only reasonable size for medium and large organizations, everyone reached for it, driving Class B toward exhaustion. The natural workaround — handing out many stitched-together Class C blocks — created a second crisis: each Class C had to be advertised as its own route, so backbone routing tables exploded toward the memory limits of the routers themselves.

Figure 3.3: The classful lose-lose dilemma

The solution, introduced in 1993 and formalized in RFC 1519, was CIDR — Classless Inter-Domain Routing. CIDR abandons the rigid A/B/C boundaries and lets the network/host split fall at any bit position, expressed as a prefix length like /20 or /27. This delivered two decisive benefits:

1. Right-sized allocations. The org needing 2,000 hosts gets a /21 (2,046 usable hosts) instead of a full /16 — cutting waste from ~63,000 addresses to a few dozen.
2. Route aggregation (supernetting). Contiguous prefixes summarize into one entry: 192.168.0.0/24 through 192.168.3.0/24 collapse into a single 192.168.0.0/22, slowing routing-table growth.

Working alongside CIDR is VLSM (Variable-Length Subnet Masking), which lets a single network be subdivided into subnets of different sizes — a user LAN gets a /24 while a point-to-point WAN link gets a tiny /30 — instead of forcing one fixed mask everywhere.

Figure 3.4: How CIDR and VLSM answer the classful failures