Ethernet (IEEE 802.3) — CSMA - CD, frame format: Question bank

Before we start, a one-line refresher of the two clocks so every reveal below is unambiguous:

True or false — justify

Every answer below must give a reason, never a bare yes/no.

TF1. "CSMA/CD eliminates collisions entirely."

False. Carrier sense only prevents you from starting on a busy wire; two idle-sensing stations can still start within one propagation window and collide. CSMA/CD detects and recovers, it does not prevent.

TF2. "On a full-duplex switched link, CSMA/CD is still running."

False. Full-duplex means separate send/receive paths with no shared medium, so collisions are impossible and CSMA/CD is disabled. It only applies to half-duplex shared Ethernet — see Switched vs Shared Ethernet.

TF3. "Doubling the cable length forces a larger minimum frame size."

True.

L_{min} = 2 T_{p} B

and

T_{p} = d / v

, so doubling

d

doubles

T_{p}

and doubles

L_{min}

— the sender must stay on-air longer for the longer round trip.

TF4. "The preamble is part of the 64-byte minimum frame."

False. The 64-byte minimum is measured from Destination MAC through FCS. Preamble (7B) + SFD (1B) are physical-layer sync bits sent before the counted frame.

TF5. "The jam signal carries the sender's data so the receiver can recover it."

False. The 32-bit jam is deliberate noise whose only job is to make the collision long enough that every station is certain to notice; the collided data is discarded, not recovered.

TF6. "A station that senses the medium idle can transmit with zero collision risk."

False. "Idle" only means no signal has reached it yet. A far station may already be transmitting, its signal still in flight — this in-flight window is the entire reason collisions exist.

TF7. "Binary exponential backoff guarantees the two colliding stations never collide again."

False. They pick

K

randomly from the same set, so they can draw the same

K

and re-collide; doubling the window only lowers that probability, never zeroes it.

TF8. "A collision corrupts the frame, so the FCS (CRC) is what tells the sender a collision happened."

False. The sender detects a collision by hearing abnormal signal energy on the wire while transmitting, not by CRC. CRC is checked by the receiver for ordinary bit errors — see CRC and Error Detection.

Spot the error

Each statement contains one flaw. Name it.

SE1. "Minimum data field is 46 bytes because 46 is half of a typical MTU."

The 46 comes from arithmetic:

6 + 6 + 2 + data + 4 = 64 \Rightarrow data = 46

. It backs out the CSMA/CD 64-byte floor, nothing to do with MTU halving.

SE2. "Length/Type = 0x0500 means EtherType IPv4."

0 x 0500 = 1280 \leq 1500

, so this value is a length, not a type. EtherType requires the field

\geq 1536

(0x0600); the 1500–1536 gap exists precisely to remove this ambiguity.

SE3. "After the 3rd collision, backoff K is chosen from {0,1,2,3}."

Wrong range. After the

n

-th collision

K \in {0, \dots, 2^{n} - 1}

, so

n = 3

gives

{0, \dots, 7}

. The set

{0, 1, 2, 3}

n = 2

SE4. "A station gives up after 16 collisions, and the window keeps doubling all 16 times."

The window stops growing at

n = 10

(

K \in {0, \dots, 1023}

); attempts 11–16 reuse that same window. Give-up is at 16 attempts, but doubling caps at 10.

SE5. "Because

T_{t x} \geq 2 T_{p}

, we should make frames as short as possible."

Backwards. Short frames violate the constraint; the inequality forces a minimum length so detection works. Efficiency wants frames longer than the minimum, not shorter.

SE6. "Preamble is 10101011 repeated to synchronize the clock."

The Preamble is 10101010×7 (alternating). The pattern ending in ...11 is the 1-byte SFD (10101011), whose final two consecutive 1s mark the exact frame start.

SE7. "The maximum frame size is 1500 bytes."

1500 is the max data field. The maximum frame (DA→FCS) is

6 + 6 + 2 + 1500 + 4 = 1518

bytes.

Why questions

WY1. Why must a station keep listening while it transmits, instead of checking after?

A late check means the frame is already fully sent and gone; the sender would wrongly believe it succeeded. Detecting during transmission is what lets it abort and retry, which is the reason for the

T_{t x} \geq 2 T_{p}

rule.

WY2. Why is the backoff randomized rather than a fixed wait?

If both stations waited an identical fixed time, they would retry simultaneously and collide forever. Random waits break the symmetry so they usually retry at different moments.

WY3. Why does the window double after each collision instead of growing by a fixed amount?

Repeated collisions signal a more crowded medium; a geometrically larger window spreads retries over more slots, cutting the chance two stations pick the same slot much faster than linear growth would.

WY4. Why does wireless (802.11) use collision avoidance (CSMA-CA) instead of detection?

A radio can't reliably hear a faint remote signal while its own powerful transmit signal drowns the receiver, so "listen-while-sending" fails. CSMA-CA sidesteps this by reserving the channel before sending rather than detecting mid-transmission.

WY5. Why does CSMA improve on plain ALOHA and Slotted ALOHA?

ALOHA transmits without listening, so collisions are frequent. Carrier sensing means you only start when the medium already sounds idle, eliminating the many collisions caused by starting on top of an ongoing transmission.

WY6. Why is the slot time defined as

2 T_{p}

(the round trip) and not

T_{p}

A collision is only guaranteed detectable within one full round trip: signal out plus garble back. Backing off in units of this worst-case round trip ensures a retrying station waits at least long enough for the medium to truly clear.

WY7. Why does giving up happen after a fixed 16 attempts rather than trying forever?

Endless retries would trap the station and swamp the wire; after 16 failures the network is deemed too congested or broken, so the error is escalated to the upper layer to handle.

Edge cases

EC1. What happens to CSMA/CD if the cable is made so long that

L_{min}

would exceed the maximum frame (1518 B)?

Detection can no longer be guaranteed for legal frames, so such a segment is illegal. This is exactly why Ethernet caps segment length (and hub diameter) — the geometry must keep

2 T_{p} B \leq

a sendable frame.

EC2. A station has only 20 bytes of real data to send. What goes on the wire?

The data field is padded with filler up to 46 bytes so the frame reaches the 64-byte minimum; the receiver uses the Length field to know only 20 bytes are real payload.

EC3. Two stations collide, and both draw

K = 0

on backoff. What happens?

Both wait zero slots and retry immediately, colliding again. This is allowed and expected occasionally; the next collision raises

n

, widening the window and reducing the chance of a repeat.

EC4. A collision is detected on the very last bit of a minimum-size frame. Is the design still safe?

Yes — that boundary is the whole point.

T_{t x} = 2 T_{p}

is set so the sender is still transmitting exactly until the worst-case echo returns; detecting on the final bit is the tightest valid case.

EC5. Only one station is connected to a shared hub. Can a collision ever occur?

With a single transmitter no two signals overlap, so no collision arises — but CSMA/CD still runs, since half-duplex shared mode doesn't know it's alone.

EC6. The medium stays busy essentially forever (a jabbering station). What does a well-behaved station do?

It keeps sensing and defers indefinitely without transmitting — carrier sense means it never barges in. It cannot force access; the fault must be fixed at the physical/hardware level.

EC7. The Destination MAC is all ones (FF:FF:FF:FF:FF:FF). How is this frame treated?

That is the broadcast address, so every station on the segment accepts and processes it — see MAC Addresses. It still obeys the same CSMA/CD sensing and minimum-size rules.

EC8. A frame arrives that is shorter than 64 bytes (a "runt"). What does the receiver conclude?

It treats the runt as a collision fragment or damaged frame and discards it — legal frames can never be below 64 bytes, so anything shorter is evidence of a collision or fault.

Recall One-line self-check

If a friend says "we should shrink frames to detect collisions faster," what's your one-sentence rebuttal? ::: Shorter frames break detection — the sender must stay on-air for a full round trip ( $T_{t x} \geq 2 T_{p}$ ), so there is a minimum, not a target-to-minimize.