In 1988, defense contractor Halliburton owned 16.7 million public IP addresses—more than the total allocated infrastructure of China at the time—simply because they asked for a block early on.

This massive imbalance was not a rogue policy decisions; it was a core feature of the original internet architecture known as Classful Networking. When engineers drew up the foundational rules for network routing, they could not envision a world with billions of connected smartphones, laptops, and smart refrigerators.

To understand modern networking and how traffic traverses the web today, you need to understand this original class system, the mechanics behind it, and why engineers eventually had to dismantle it.

The 1981 Master Blueprint (RFC 791)

In September 1981, the Internet Engineering Task Force (IETF) published RFC 791, laying down the official specification for the Internet Protocol Version 4 (IPv4). In an era when enterprise router hardware operated with kilobytes of memory and low clock speeds, routing packets efficiently required a massive logical shortcut.

To achieve this, the architecture divided every 32-bit IPv4 address into two functional components: a Network ID and a Host ID.

Network ID: Operates like a postal ZIP code, directing traffic to a specific physical or logical network infrastructure.
Host ID: Operates like a street address number, pointing to the exact device or interface connected within that network.

To prevent routers from having to perform deep lookups across huge tables, designers categorized addresses into discrete classes based on the very first bits of the address—known as the leading bits.

By inspecting the highest-order bits of the first octet, a 1980s router could instantly determine where the network portion ended and where the host portion began.

Breaking Down Class A, B, and C

An IPv4 address consists of four 8-bit octets (32 bits total), written in dotted-decimal format like 192.168.1.1. This provides a maximum address space of $2^{32}$, or approximately 4.29 billion addresses. RFC 791 sliced this pie into three primary unicast classes based on organizational scale.

Class A: The Megacorporation Tier

Designed for massive national networks, Class A addresses reserve the first bit as a mandatory 0.

First Octet Range: 1 to 126 (Binary: 00000001 to 01111110)
Network Portion: 8 bits (Default mask: 255.0.0.0 or /8)
Host Portion: 24 bits
Total Class A Networks: 126
Usable Hosts per Network: $2^{24} - 2 = 16,777,214$

The subtraction of 2 applies across all host calculations: the all-zeros host address represents the network identifier itself, while the all-ones address serves as the subnet broadcast address.

Early tech pioneers and institutions landed massive Class A blocks for free. Apple bagged 17.0.0.0/8, Ford took 19.0.0.0/8, MIT received 18.0.0.0/8, and Stanford secured 36.0.0.0/8. MIT alone held more public IPv4 space than entire industrialized nations for decades.

Class B: Mid-Sized Universities and Enterprises

Class B addresses targeted mid-sized organizations, such as regional universities, healthcare systems, and large corporate campuses. The first two bits are hardcoded to 10.

First Octet Range: 128 to 191 (Binary: 10000000 to 10111111)
Network Portion: 16 bits (Default mask: 255.255.0.0 or /16)
Host Portion: 16 bits
Total Class B Networks: 16,384
Usable Hosts per Network: $2^{16} - 2 = 65,534$

Class B struck a middle ground, but as business computing exploded in the late 1980s, these blocks became the most contested property on the internet.

Class C: Small Local Networks

Class C addresses were allocated to small businesses and local area networks (LANs). The leading bit pattern is fixed to 110.

First Octet Range: 192 to 223 (Binary: 11000000 to 11011111)
Network Portion: 24 bits (Default mask: 255.255.255.0 or /24)
Host Portion: 8 bits
Total Class C Networks: 2,097,152
Usable Hosts per Network: $2^8 - 2 = 254$

While 254 addresses suited a small office environment, mid-sized companies with 500 employees found a single Class C insufficient, forcing them into a structural dilemma.

These three tiers covered standard device-to-device communication, but two additional classes were established for specialized purposes.

Class D and E: Multicast and the Experimental Void

Beyond unicast addressing, the IETF set aside two entire classes that did not use the standard network/host split.

Class D: Multicast Traffic

Class D addresses begin with the four-bit binary sequence 1110.

First Octet Range: 224 to 239 (224.0.0.0 to 239.255.255.255)
Designated Function: IP Multicasting

Class D networks do not contain host bits. Instead of addressing a single physical NIC, a Class D address represents a dynamic group of hosts listening for the same stream. Applications include live video feeds, financial stock market tickers, and routing protocols like OSPF sending topology updates to neighboring routers.

Class E: The Experimental Lockout

Class E addresses begin with 1111.

First Octet Range: 240 to 255 (240.0.0.0 to 255.255.255.255)
Designated Function: Reserved for IETF research and future development

Class E consumes 268 million addresses—roughly 6.25% of the total IPv4 space. Because TCP/IP stack implementations in nearly all operating systems were programmed to discard Class E traffic as invalid, this vast block remains unusable on the public internet today, despite total IPv4 exhaustion.

Outside of these structural class definitions, specific address blocks inside these ranges were permanently set aside for non-routable functions.

Reserved Addresses and Special Networks

Not every address within Classes A, B, and C can be assigned to a server or workstation on the public web. Key exceptions include:

Loopback Addresses (127.0.0.0/8): The entire 127 block—historically inside Class A—was carved out for local host diagnostic software (127.0.0.1). This single engineering decision effectively sacrificed 16.7 million addresses solely so a machine could ping itself.
Private Address Space (RFC 1918): Released in February 1996, RFC 1918 reserved specific ranges within the legacy classes for internal corporate networks, inaccessible from the public internet without Network Address Translation (NAT):
- 10.0.0.0/8: One full Class A network (16.7 million addresses)
- 172.16.0.0/12: 16 contiguous Class B networks
- 192.168.0.0/16: 256 contiguous Class C networks
Link-Local / APIPA (169.254.0.0/16): Reserved for automatic address allocation when a device cannot reach a DHCP server, allowing local peer-to-peer communication on a single switch segment.

Understanding these rigid allocations brings us to the structural crisis that hit the global internet infrastructure in the early 1990s.

The Inefficiency Crisis: Why Classful Routing Died

By 1992, the internet was suffering from a classic “Goldilocks Problem.”

Class C blocks (254 hosts) were too small for most growing organizations. Class A blocks were all gone. That left Class B blocks (65,534 hosts) as the only viable option for any enterprise with more than 200 devices.

Consider an organization needing addresses for 1,000 workstations in 1992:

Assigning 4 Class C blocks created four distinct routes in core internet routing tables, causing routing table bloat that threatened to crash global routers.
Assigning 1 Class B block gave the company 65,534 addresses. The company used 1,000 addresses and left 64,534 addresses completely stranded and unusable by anyone else.

By late 1992, over 60% of all Class B space had been allocated, and projections showed the global internet running completely out of assignable IPv4 space by 1996.