Free lesson · Computer Networking

Ethernet The Foundation Of Local Networks

Easy10 min readnetworkingcomputer-networkstcp-ip

Why Ethernet Dominates

If you have ever plugged a cable into a laptop, connected a server to a rack switch, or set up a home router, you have used Ethernet. It is the dominant technology for local area networking and has been for over four decades. In data centers, in offices, in homes - Ethernet is the standard for wired local network connectivity.

Ethernet operates at Layer 2 (Data Link) and Layer 1 (Physical) of the OSI model. The Physical layer defines the cables, connectors, and signaling. The Data Link layer defines how data is organized into frames, how devices are identified using MAC addresses, and how multiple devices share access to the same network medium.

The original Ethernet was invented at Xerox PARC in the early 1970s by Robert Metcalfe, who named it after the "luminiferous ether" - the hypothetical medium that nineteenth-century physicists believed filled all of space. The first version ran at 2.94 Mbps over a shared coaxial cable. Today, Ethernet runs at speeds up to 400 Gbps and beyond, over copper and fiber optic cables, with dedicated point-to-point connections instead of shared media. The name and the core framing structure survived; almost everything else evolved. The IEEE standardized Ethernet as IEEE 802.3 in 1983, and the 802.3 working group continues to publish new Ethernet standards for higher speeds and new media types.

What makes Ethernet so dominant is not any single technical advantage - it is a combination of simplicity, backward compatibility, and ecosystem momentum that created a self-reinforcing cycle of adoption. Ethernet has continuously evolved to faster speeds while maintaining the same basic frame format. A Gigabit Ethernet switch can communicate with a 10 Gigabit Ethernet switch because the frame structure is the same - only the physical layer changes. This backward compatibility protected the enormous investment companies made in Ethernet infrastructure, and it is why no competing LAN technology has ever displaced it.

The Ethernet Frame - Data Link Layer Structure

The Ethernet frame is the fundamental unit of data at Layer 2. Every piece of data that crosses an Ethernet network - whether it is a web request, a DNS query, or a video stream - is wrapped in an Ethernet frame for local delivery.

The frame has a precisely defined structure:

graph LR
    subgraph Frame["Ethernet Frame"]
        direction LR
        PRE["Preamble<br/>7 bytes"]
        SFD["SFD<br/>1 byte"]
        DMAC["Dest MAC<br/>6 bytes"]
        SMAC["Src MAC<br/>6 bytes"]
        TYPE["EtherType<br/>2 bytes"]
        DATA["Payload<br/>46-1500 bytes"]
        FCS["FCS<br/>4 bytes"]
    end

    PRE --> SFD --> DMAC --> SMAC --> TYPE --> DATA --> FCS

    style PRE fill:#c4b5fd,stroke:#a78bfa,color:#000
    style SFD fill:#c4b5fd,stroke:#a78bfa,color:#000
    style DMAC fill:#7c3aed,stroke:#6d28d9,color:#fff
    style SMAC fill:#7c3aed,stroke:#6d28d9,color:#fff
    style TYPE fill:#8b5cf6,stroke:#7c3aed,color:#fff
    style DATA fill:#4f46e5,stroke:#4338ca,color:#fff
    style FCS fill:#a78bfa,stroke:#8b5cf6,color:#fff

This diagram shows the structure of an Ethernet frame as a horizontal sequence of fields. From left to right: the Preamble is 7 bytes for synchronization, the Start Frame Delimiter is 1 byte, the Destination MAC Address is 6 bytes, the Source MAC Address is 6 bytes, the EtherType field is 2 bytes identifying the payload protocol, the Payload contains 46 to 1500 bytes of data, and the Frame Check Sequence is 4 bytes for error detection.

Key Details About the Frame

Minimum frame size is 64 bytes (excluding preamble and SFD). If the payload is less than 46 bytes, padding bytes (zeros) are added to meet the minimum. This minimum was originally needed for collision detection in shared-media Ethernet - a frame needed to be long enough that a collision could be detected before the sender finished transmitting. Although collisions no longer occur in modern switched Ethernet, the minimum frame size remains part of the standard for backward compatibility, and all Ethernet implementations still enforce it.

Maximum payload size is 1500 bytes by default - this is the standard Maximum Transmission Unit (MTU) for Ethernet. If the upper layer hands down more than 1500 bytes of data, it must be split into multiple frames. Some networks support jumbo frames with payloads up to 9000 bytes, which reduce header overhead for large data transfers but require all devices on the network segment to support them.

The FCS field contains a CRC-32 checksum computed over the destination MAC, source MAC, EtherType, and payload fields. The receiving NIC computes the same checksum over the received data and compares it to the FCS value. If they do not match, the frame is silently discarded - there is no retransmission at Layer 2. Reliability is the responsibility of upper layers (specifically TCP at Layer 4).

Important: Ethernet detects errors but does not correct them. If a frame arrives with a bad FCS, the receiving device drops it without notifying the sender. This is a deliberate design choice - keeping Layer 2 simple and fast, and leaving reliability to TCP. This is an important example of the "end-to-end principle" in network design: put complex logic at the endpoints, keep the network itself simple.

CSMA/CD - How Old Ethernet Managed Collisions

Early Ethernet networks used a shared medium - a single coaxial cable that every device was physically connected to. This created a fundamental problem: if two devices transmit at the same time, their signals collide and both transmissions are corrupted.

Ethernet solved this with CSMA/CD - Carrier Sense Multiple Access with Collision Detection. The name describes the entire algorithm:

1. Carrier Sense: Before transmitting, the device listens to the cable. If it detects another signal (carrier), it waits.
2. Multiple Access: All devices share the same cable and have equal access to it. There is no central controller deciding who transmits when.
3. Collision Detection: If the device starts transmitting and detects a collision (the signal on the wire does not match what it sent), it immediately stops, sends a jam signal to notify all devices, and waits a random amount of time before trying again.

The random wait time uses an algorithm called binary exponential backoff: after the first collision, the device waits a random time between 0 and 1 time slot. After the second collision, between 0 and 3 time slots. After the third, between 0 and 7. Each successive collision doubles the maximum wait, up to a cap. This randomization ensures that colliding devices do not keep retrying at the same instant.

Tip: CSMA/CD is historically important but largely irrelevant in modern networks. Today's Ethernet uses full-duplex, point-to-point switched connections. Each device has a dedicated cable to a switch port - there is no shared medium and therefore no collisions. Understanding CSMA/CD matters for two reasons: it explains why hubs were replaced by switches (eliminating collisions), and it explains the minimum frame size (collision detection required frames of a certain duration).

Full-Duplex Ethernet - The Modern Reality

Modern Ethernet eliminates collisions entirely through two changes:

Switches instead of hubs. Each device connects to a dedicated switch port via its own cable. The switch creates a separate collision domain for each port. Two devices can transmit simultaneously because their signals travel on separate cables to separate switch ports.

Full-duplex communication. On a point-to-point switched connection, the device can transmit and receive simultaneously. The cable has separate wire pairs for sending and receiving. There is no need to listen before transmitting because there is no shared medium and no possibility of collision.

Full-duplex Gigabit Ethernet provides 1 Gbps in each direction simultaneously - a total of 2 Gbps of aggregate bandwidth per link. This is a significant improvement over old half-duplex shared Ethernet, where all devices competed for a single shared bandwidth.

The transition from shared-media CSMA/CD Ethernet to switched full-duplex Ethernet is one of the most significant performance improvements in networking history. It happened gradually during the late 1990s and early 2000s as switches became affordable enough to replace hubs. By the mid-2000s, shared-media Ethernet had effectively vanished from production networks. Today, every Ethernet connection you encounter - in a data center, in an office, in a home - is switched and full-duplex.

This matters because it means the theoretical performance characteristics you see in specifications (1 Gbps, 10 Gbps, 25 Gbps) represent actual usable bandwidth per link, not shared bandwidth divided among all devices. Each device gets the full link speed on its dedicated connection to the switch. When people talk about "a Gigabit network," they mean each connected device has a dedicated 1 Gbps full-duplex link, not that all devices share 1 Gbps among them.

The Evolution of Ethernet Speed

Ethernet has scaled from 10 Mbps to 400 Gbps over four decades while maintaining the same basic frame format:

The naming convention follows a consistent pattern: the number before BASE indicates speed in Mbps or Gbps, BASE means baseband signaling (the entire cable bandwidth is used for a single channel, as opposed to broadband which divides the cable into multiple channels), and the suffix indicates the medium or distance - T for twisted-pair copper, SR for short-range multi-mode fiber, LR for long-range single-mode fiber, and CR for copper direct-attach cables used in data centers.

Each speed increase was achieved through changes at the Physical layer - better cables, more sophisticated encoding, more wire pairs or fiber strands used in parallel - while the Data Link layer frame format remained essentially unchanged. A 400 Gbps Ethernet frame has the same structure as a 10 Mbps Ethernet frame. This backward compatibility is a remarkable engineering achievement.

graph TD
    subgraph Evolution["Ethernet Speed Evolution"]
        direction TB
        E10["1990: 10 Mbps<br/>10BASE-T"]
        E100["1995: 100 Mbps<br/>Fast Ethernet"]
        E1G["1999: 1 Gbps<br/>Gigabit Ethernet"]
        E10G["2006: 10 Gbps<br/>10G Ethernet"]
        E25G["2016: 25 Gbps"]
        E100G["2010s: 100 Gbps"]
        E400G["2018+: 400 Gbps"]
    end

    E10 --> E100
    E100 --> E1G
    E1G --> E10G
    E10G --> E25G
    E25G --> E100G
    E100G --> E400G

    style E10 fill:#c4b5fd,stroke:#a78bfa,color:#000
    style E100 fill:#a78bfa,stroke:#8b5cf6,color:#fff
    style E1G fill:#8b5cf6,stroke:#7c3aed,color:#fff
    style E10G fill:#7c3aed,stroke:#6d28d9,color:#fff
    style E25G fill:#6366f1,stroke:#4f46e5,color:#fff
    style E100G fill:#4f46e5,stroke:#4338ca,color:#fff
    style E400G fill:#4338ca,stroke:#3730a3,color:#fff

This diagram shows the evolution of Ethernet speeds over time, arranged vertically from oldest to newest. Starting in 1990 with 10 Mbps 10BASE-T, then 1995 with 100 Mbps Fast Ethernet, 1999 with 1 Gbps Gigabit Ethernet, 2006 with 10 Gbps, 2016 with 25 Gbps, the 2010s with 100 Gbps, and 2018 onward with 400 Gbps. Each step represents a 2.5x to 10x speed increase while maintaining the same basic frame format.

Where Ethernet Lives in the Modern Network

Understanding where Ethernet operates helps clarify what it does and does not do:

Ethernet operates within a single network segment - from a device to a switch, or between two switches in the same local network. It does not route across networks. When a packet needs to leave the local network and travel to a different network, a router strips the Ethernet frame, reads the IP packet inside, makes a routing decision, and wraps the IP packet in a new Ethernet frame (or a different Layer 2 protocol) for the next segment of the journey.

Ethernet is not the only Layer 2 technology, but it is overwhelmingly the most common one for wired LANs. Other Layer 2 technologies include Wi-Fi (IEEE 802.11) for wireless, PPP (Point-to-Point Protocol) for some WAN links, and MPLS (Multiprotocol Label Switching) for carrier networks. In practice, you can assume that any wired local network you encounter uses Ethernet unless you have specific reason to think otherwise.

In the cloud context, Ethernet is equally dominant. When you provision a virtual machine in AWS, Azure, or GCP, the virtual network interface presented to your VM emulates an Ethernet NIC. The VM sends and receives Ethernet frames, even though the underlying physical infrastructure may use sophisticated overlay networking. The Ethernet abstraction is so universal that even virtualized environments maintain it as the standard interface between the operating system and the network.

Interview: Ethernet itself is rarely the subject of System Design interview questions. But the concepts embedded in Ethernet - frame-level error detection, MAC addressing, MTU limits, full-duplex communication - show up in discussions about:

• MTU and fragmentation: "What happens when a server tries to send a 9000-byte response over a network with a 1500-byte MTU?" Understanding that the IP layer must fragment the data into multiple Ethernet frames explains performance implications.
• Network segmentation: "How would you isolate traffic between different services in a data center?" VLANs (covered in Topic 5) are an Ethernet-level mechanism for logical segmentation.
• Full-duplex vs half-duplex: Understanding that modern switched Ethernet is full-duplex explains why modern LANs are fast and collision-free.

What Comes Next

The next topic examines MAC addresses - the hardware-level identity that Ethernet uses to deliver frames to the correct device on a local network. You have seen that every Ethernet frame carries a source MAC and destination MAC. The next topic explains what those addresses actually are, how they are assigned, and how they differ from IP addresses.

Exercise: Starter: Check your device's Ethernet or Wi-Fi MAC address. On Mac/Linux, run ifconfig or ip link show in a terminal. On Windows, run ipconfig /all in Command Prompt. Find the "Physical Address" or "ether" field. What format is it in? How many groups of hex digits do you see?

Intermediate: If a standard Ethernet frame has a maximum payload of 1500 bytes, and an IP header is 20 bytes, and a TCP header is 20 bytes, what is the maximum amount of application data (HTTP body, for example) that can fit in a single Ethernet frame? Now consider: if you are transferring a 1 MB file, how many Ethernet frames are needed at minimum?

Advanced: Research the trade-offs of jumbo frames (9000-byte MTU) in data center networks. Why do some data centers enable jumbo frames and others do not? What happens when a jumbo frame hits a link that only supports 1500-byte MTU? How does Path MTU Discovery (PMTUD) handle this situation?

Back to Computer Networking