Ethernet: Frames, Shared-Medium History, and Modern Full-Duplex Links

From a Shared Bus to Switches

Ethernet appeared in the mid-1970s and is today's most widely used wired LAN technology. The original Ethernet used a coaxial cable as a bus to which every node was attached: exactly the shared-medium network of the previous section. Any frame a node sent was received by all the others, and nodes coordinated their transmissions with CSMA/CD.

In the 1990s, Ethernet moved to twisted-pair cables in a star topology with a hub at the center. A hub is a physical-layer device that repeats the bits received on one interface out of all its other interfaces, so the network is still logically one shared "bus", and collisions still happen.

Today, the center of the star is almost always a switch. A switch is a link-layer device that stores and forwards whole frames, and it forwards a frame only out of the interface leading to its destination.

Non-Shared-Medium Networks

In a non-shared-medium network, every station connects directly to a switch, and the switch forwards the frames. Such networks generally use full-duplex communication.

Suppose A sends data whose destination address is B. The switch forwards the data to station B.

Because each link connects just one node and one switch interface, and both directions can transmit at the same time, collisions cannot happen on the link at all, so CSMA/CD is no longer needed. While A sends to B, C can send to D, and the switch forwards both frames independently.

The Ethernet Frame

Whatever the physical medium and the rate, the Ethernet frame format has stayed essentially the same for decades:

Ethernet offers a connectionless, unreliable service: there is no handshake before sending, and frames that fail the CRC check are simply dropped, with neither acknowledgment nor retransmission. When reliable transfer is needed, TCP in a higher layer detects the loss and retransmits.

CSMA/CD and Binary Exponential Backoff

On shared-medium Ethernet, an adapter sends frames following the CSMA/CD rules:

1. The adapter receives a datagram from the network layer and encapsulates it in a frame.
2. It senses the channel. If the channel is idle, it starts transmitting; if busy, it waits until the channel becomes idle and then transmits.
3. It keeps monitoring the channel while transmitting. If it sends the whole frame without detecting another signal, the frame has been sent successfully.
4. If it detects a collision, it stops transmitting at once and sends a 48-bit jam signal to make sure every node knows about the collision.
5. After aborting, it enters binary exponential backoff: after the $n$-th collision it chooses $K$ at random from $\{0, 1, \dots, 2^m - 1\}$, where $m = \min(n, 10)$, waits $K \times 512$ bit times, and returns to step 2. It gives up after 16 consecutive failures.

More collisions suggest more competing nodes, so the range of waiting times doubles each time and quickly lowers the chance of another collision; with few competitors, the waits stay short.

The minimum frame size also comes from CSMA/CD: transmitting a minimum frame (64 bytes, or 512 bits) must take at least as long as a signal's round trip across the network, so the sender is still transmitting, and is sure to detect the collision, whenever one happens.

Ethernet Standards

There are many Ethernet standards, and their names often take the form "rate + BASE + medium": BASE means baseband transmission, T means twisted pair, and F, SX, LR and so on denote fiber.

The rate has grown by orders of magnitude, but the frame format has not changed, so the upper layers never notice the change in the underlying link. Modern switched Ethernet runs in full-duplex mode, and CSMA/CD is kept only for compatibility.

ffffffffffff 0800200a8c6d 0806 0001080006040001 0800200a8c6d c0a80105 000000000000 c0a80101