14.3.1 Evolution Of Ethernet

Ethernet was originally developed in the 1970s as a shared-medium local area network operating over coaxial cable. Early implementations employed a logical bus topology in which all devices were connected to a single transmission medium. Because the medium was shared, only one device could successfully transmit at a time.

To regulate access to the shared channel, Ethernet adopted CSMA/CD as a media-access scheme. Under this scheme, a device wishing to transmit first listened to the medium to determine whether it was idle (carrier sense). If no signal was detected, transmission began immediately. If two devices transmitted simultaneously, a collision occurred. Each transmitting device monitored the medium while transmitting and, upon detecting a collision, ceased transmission and invoked a retransmission procedure. Upon detecting a collision, a transmitting station also emitted a short jam signal (at least 32 bits) to ensure that the collision condition was recognized by all stations within the collision domain before transmission was terminated.

Retransmission was governed by the truncated binary exponential backoff algorithm. After each detected collision, a station waited for a randomly selected delay interval before attempting to retransmit. If further collisions occurred, the waiting interval increased exponentially. This mechanism reduced the probability of repeated collisions under heavy load and allowed the network to stabilize during periods of contention.

Early shared-medium Ethernet was subject to important design constraints that limited both the number of attached stations and the physical extent of the network. Although CSMA/CD provided a mechanism for managing contention, excessive numbers of active stations could drive the network toward saturation, where repeated collisions consumed a substantial fraction of available bandwidth. For practical stability, the number of attached devices was therefore constrained (historically up to 1,024 stations per collision domain under repeater-based architectures). In addition, the maximum physical diameter of the network was tightly limited by collision-detection timing requirements. A transmitting station had to remain active long enough to detect a collision caused by the most distant station on the segment. This requirement determined both the maximum cable length and the minimum Ethernet frame size (64 octets). These timing and distance constraints played a critical role in shaping early Ethernet physical-layer design and will be revisited in Section 14.3.4 when considering the evolution of Ethernet’s PHY.

The original Ethernet standard operated at 10 Mbps. Because of the overhead associated with contention and collision recovery, effective throughput was significantly lower under heavy load conditions. Subsequent developments increased the data rate to 100 Mbps (Fast Ethernet), 1 Gbps (Gigabit Ethernet), 10 Gbps, and into the multi-hundred-gigabit range for data center and backbone applications, while preserving backward compatibility at the frame level.

A fundamental architectural shift occurred with the introduction of Ethernet switches and full-duplex operation. Rather than sharing a single bus, modern Ethernet networks connect each device to a dedicated switch port using twisted-pair or fiber-optic links. In full-duplex mode, transmission and reception occur simultaneously over separate pairs or fibers, and collisions are no longer possible. Consequently, CSMA/CD is disabled in modern switched Ethernet deployments.

Although contemporary Ethernet no longer relies on shared-medium contention, understanding CSMA/CD remains important for appreciating the design constraints that shaped the Ethernet frame format and minimum frame size. The protocol’s historical evolution also explains terminology such as “collision domain” and “segment,” which persist in network design discussions.

Although Ethernet’s access method and physical architecture have evolved significantly, its Data Link Layer frame format has remained remarkably stable across generations. We therefore now examine the Ethernet frame and the Media Access Control (MAC) functions that govern how frames are structured, addressed, and validated at the Data Link Layer.