ELEC3506

MAC and LANsLecture 415 min

Ethernet Standards, Full Duplex and Switching

How the 802.3 family scales from thick coax to 10-gigabit fibre on one shared frame format, and what a switch actually does to a collision domain.

By the end of this page you should be able to

  • Read an Ethernet standard name like 10BASE-T and say what each part means
  • List the four common 10-Gigabit implementations and their encodings
  • Explain why full-duplex Ethernet has no need for CSMA/CD
  • Distinguish cut-through switching from store-and-forward switching
  • Explain what a switch does to a network's collision domains, and what it leaves untouched

The idea

Every 802.3 standard, from the original coaxial cable to 10-gigabit fibre, shares the same MAC protocol and the same frame format covered on the last two pages. What changes between standards is the physical layer underneath it — the medium, the speed, and increasingly, whether contention happens at all.

Traditional Ethernet

How it works

Coax and twisted pair

  • 10BASE5 — thick coaxial cable, the original traditional Ethernet.
  • 10BASE2 — thin coaxial cable, same idea, cheaper and more flexible cabling.
  • 10BASE-T — twisted-pair copper, and the first standard here that supports full duplex.
  • 10BASE-F — fibre.

10BASE5 and 10BASE2 are half-duplex, bus-topology networks — every station shares one physical medium, and everything from the CSMA/CD page applies directly. 10BASE-T changes the topology as well as the cable: stations connect point-to-point to a switch rather than sharing one bus.

802.3 covers a wide speed range on top of this: 2 Mbps, 10 Mbps, 100 Mbps, 1 Gbps, 10 Gbps and 40 Gbps. Fast Ethernet (100 Mbps) includes several named variants — 100BASE-TX, -T2, -T4, -FX, -SX and -BX — and Gigabit Ethernet runs at 1 Gbps. The slides name these but do not give the media type or distance detail for each one individually, so treat the names as what is examinable here rather than their per-standard specifications.

10-Gigabit Ethernet

10-Gigabit Ethernet, IEEE 802.3ae, operates only in full duplex. There is no contention and no CSMA/CD at this speed at all.

ImplementationMediumMax lengthWiresEncoding
10GBASE-SRFibre, 850 nm300 m264B/66B
10GBASE-LRFibre, 1310 nm10 km264B/66B
10GBASE-EWFibre, 1350 nm40 km2SONET
10GBASE-X4Fibre, 1310 nm300 m – 10 km28B/10B
The four common 10-Gigabit implementations, as the lecture lists them.

Aside

Standard names for the same families

The lecture’s “10GBASE-X4” is shorthand for a family IEEE 802.3ae names explicitly: 10GBASE-LX4 and 10GBASE-CX4, both using 8B/10B encoding. On the 64B/66B side, the standard also defines 10GBASE-ER alongside -SR and -LR. Confirmed against the IEEE 802.3ae naming, not stated by name in the slides.

Full duplex removes the problem entirely

10BASE5 and 10BASE2 are half-duplex — one shared medium, one direction of traffic at a time. 10BASE-T is full duplex: two separate links between a station and its switch, one for transmitting and one for receiving. With a dedicated link in each direction, there is no need for CSMA/CD — nothing shared to contend for, nothing that can collide.

That matters more as speed increases. The same 512-bit slot time that gives 10 Mbps a 5120 m theoretical reach only allows about 51 m at 1 Gbps, because the same 512 bits leave the wire faster at a higher rate, shrinking the propagation-delay budget that keeps a half-duplex network safe. Full duplex sidesteps the problem rather than shrinking a shared medium’s usable distance further: once every station has a dedicated line in each direction, the question of how far a collision can travel does not arise.

Check yourself

Why does 1 Gbps Ethernet not use CSMA/CD at all, when 10 Mbps does?

Because full duplex removes the shared medium entirely. There is nothing to collide with once every station has its own transmit and receive line, so higher speeds sidestep the shrinking-collision-domain problem that half duplex would otherwise create.

Switching

A switch is a multiple-port bridge. Each port runs its own separate Ethernet segment — its own “spoke” — so nodes attached to different ports never contend with each other for the medium.

Unbridged EthernetBridged (switched) Ethernet
CapacityTotal capacity shared across every stationEach segment gets independent bandwidth
Collision domainOne, shared by everyoneOne per segment
Splitting a shared bus into switched segments is what raises effective bandwidth — nobody is competing with stations they used to share a wire with.

Where marks get lost

A switch shrinks collision domains, not broadcast domains

Splitting a network with switches instead of hubs solves collisions — every port is its own collision domain. It does not touch broadcast traffic. A broadcast frame still reaches every port on every switch in the network, because switching says nothing about which broadcast domain a port belongs to. That is a separate problem, and it is what VLANs solve, on the next page.

Aside

Not in the lecture: cut-through vs store-and-forward

The slides do not cover switching modes at all. This is standard material and it is examinable, so here it is.

A switch has to decide when to start forwarding a frame out the far port, relative to how much of the frame has arrived.

Store-and-forward waits for the entire frame to arrive, including the CRC, before forwarding any of it. It checks the CRC first, and only forwards a frame that passes — a corrupted frame is dropped, never forwarded. The cost is latency: the switch cannot start sending until the whole frame, potentially 1500 bytes of data, has fully arrived.

Cut-through starts forwarding as soon as it has read enough of the frame to know where it is going — just the destination address, the first bytes after the preamble and SFD. It has not seen the CRC yet, so it forwards without knowing whether the frame is actually intact. Lower latency, at the cost of occasionally forwarding a corrupted frame onward.

In the exam

  • Standard naming pattern. Speed, then signalling, then medium — 10BASE-T is 10 Mbps, baseband, twisted pair.
  • 10BASE5 and 10BASE2 are half-duplex bus. 10BASE-T and 10BASE-F support full duplex.
  • 10-Gigabit Ethernet is full-duplex only. No contention, no CSMA/CD, per IEEE 802.3ae.
  • 10GBASE-SR/-LR/-EW use 64B/66B or SONET framing; 10GBASE-X4 uses 8B/10B. Match encoding to implementation, not to speed.
  • Full duplex removes the need for CSMA/CD, because there is no shared medium left to contend for.
  • A switch is a multi-port bridge. Every port is its own collision domain, which is why switching raises effective bandwidth over a shared bus.
  • Switching does not shrink broadcast domains. That distinction is worth a mark on its own, and it is what motivates VLANs.
  • Cut-through vs store-and-forward is not lecture content, but it is examinable: cut-through trades certainty for speed, store-and-forward trades speed for certainty.

Check yourself

  1. Why does 10-Gigabit Ethernet (802.3ae) have no CSMA/CD?
  2. Which encoding do 10GBASE-SR and 10GBASE-LR use?
  3. What does a switch do to a network's collision domains?
  4. A switch reads only the destination address before it starts forwarding a frame. Which switching mode is this, and what does it trade away?
  5. Why does the theoretical collision domain shrink as line rate increases, and how does full duplex sidestep the problem?