CCNA online course

CSMA/CD Explained: Carrier Sense Multiple Access with Collision Detection

Learn how CSMA/CD manages shared half-duplex Ethernet, including collisions, jam signals, binary exponential backoff, slot time, errors, and duplex troubleshooting.

CSMA/CD stands for Carrier Sense Multiple Access with Collision Detection. It is the Ethernet media-access method used on shared, half-duplex segments. Its purpose is to coordinate access to one common transmission medium when several stations may try to send at the same time.

CSMA/CD does not prevent every simultaneous transmission in advance. Instead, stations listen before transmitting, detect collisions while transmitting, stop when a collision occurs, and retry after waiting for a randomized interval.

Prerequisites and Core Terms

You should be familiar with basic Ethernet frames and MAC addresses, the Physical and Data Link layers of the OSI reference model, network media, hubs, bridges, switches, duplex, and auto-negotiation.

Shared-Media Ethernet

In legacy Ethernet, multiple devices used one physical medium. Examples include coaxial bus Ethernet and Ethernet devices connected through a hub. A hub is a Layer 1 device: it repeats incoming electrical signals out other ports instead of making forwarding decisions.

All stations attached to a hub share one collision domain. Only one station can successfully transmit at a time on the shared half-duplex medium. If two stations transmit during overlapping signal intervals, their signals interfere and the frames are corrupted.

A switch works differently. A switch normally gives each port its own collision domain and can support full-duplex communication between an endpoint and the switch. A switch can still participate in a half-duplex link if configured or negotiated that way, but a normal modern switch-to-endpoint link is full duplex.

How CSMA/CD Works

Why Two Stations Can Collide

Imagine two hosts connected to a hub. Both listen and hear what appears to be an idle medium. They begin transmitting almost simultaneously. This is possible because a signal takes time to propagate along the medium. Each host may start before it can detect the other host's signal.

The signals meet and interfere, corrupting both frames. Each transmitting station detects a mismatch between the signal it expects and the activity on the medium. The stations stop sending their normal frames and transmit jam signals so other stations also recognize the collision.

Collisions are expected behavior on a shared half-duplex Ethernet design. A small number of collisions is not automatically a fault. High collision rates, late collisions, or degraded throughput can indicate excessive contention, an incorrect duplex setting, or a physical design problem.

Jam Signals and Collision Handling

A jam signal is a deliberate signal sent after collision detection. Its purpose is to make the collision condition long and obvious enough that all stations sharing the medium recognize it. The original collided frames are not repaired; they are discarded.

Each sender involved in the collision performs its own backoff calculation. The stations do not all wait for one fixed delay. Independent random choices reduce the chance that the same stations will immediately collide again.

Binary Exponential Backoff

Binary exponential backoff is Ethernet's randomized retry method. After a collision, a station selects a random number of slot times to wait before retrying. After consecutive collisions, the range from which that random number is selected expands, which lowers the chance of repeated collisions during heavy contention.

Conceptually, after collision number n, the station chooses a value from an increasingly large range of slot counts, subject to Ethernet's defined limits. The exact implementation details depend on the Ethernet standard, but the principle is consistent: more collisions produce a wider possible delay.

Ethernet limits the number of transmission attempts for a frame. The standard maximum is 16 attempts. If the frame still cannot be transmitted after the allowed attempts, the frame is dropped and the failure is reported as an excessive-collisions condition.

Example Backoff

1. Two hosts collide while sending.
2. Each sends a jam signal and stops.
3. Each independently selects a random number of slot times.
4. The host with the shorter delay retries first.
5. If both happen to select the same delay and collide again, the possible range expands for the next retry.

Ethernet Timing Concepts

Slot Time

A slot time is an Ethernet timing interval used for collision detection and backoff calculations. The classic Ethernet slot time is 512 bit times. It represents the time needed for a transmitting station to send the minimum frame-related detection interval while allowing a worst-case collision to propagate back to the sender.

The sender must still be transmitting when a collision from the farthest permitted point in the network returns. If the sender finished too quickly, it could incorrectly conclude that the frame was sent successfully before learning about the collision.

Minimum Frame Size

Classic Ethernet specifies a minimum frame size of 64 bytes, including the Ethernet header and FCS. If a payload is too small to produce a frame of the required size, Ethernet adds padding. This minimum keeps the transmitter active long enough to support collision detection within the network's permitted diameter and timing limits.

Network diameter is the maximum physical extent of the shared network, including the propagation time through the medium and permitted intermediary devices. Larger propagation delays require careful timing rules. Ethernet design limits, slot time, and minimum frame size work together to make collision detection reliable.

Interframe Gap

The interframe gap is the required idle period between Ethernet frame transmissions. It gives devices time to prepare for the next frame and prevents frames from being transmitted back-to-back without the defined separation.

Collision Classifications and Ethernet Errors

Collision counters and CRC/FCS counters represent different symptoms. A collision is an access event caused by overlapping transmissions on a shared half-duplex medium. A CRC or FCS error means the received frame failed its integrity check. Physical-layer problems can cause FCS errors, and a duplex mismatch can produce errors on one side while producing collision-related symptoms on the other.

Half Duplex Versus Full Duplex

A hub extends one shared collision domain because it repeats signals among all attached ports. A switch separates collision domains by port. A broadcast can still be forwarded across multiple switch ports within the same VLAN, so separating collision domains does not automatically separate broadcast domains.

Modern Relevance

Modern Ethernet networks normally use switches and full-duplex links. A host and its switch port can transmit at the same time without competing for one shared path. Therefore, CSMA/CD does not run on a correctly negotiated full-duplex switched link.

CSMA/CD remains relevant in CCNA fundamentals and in troubleshooting situations involving hubs, legacy half-duplex links, unusual shared-media environments, or duplex mismatches. Do not describe CSMA/CD as the access method used by ordinary modern full-duplex switch links.

Practical Scenarios

Two Hosts Connected Through a Hub

1. Both hosts listen and find the shared medium idle.
2. Both transmit nearly simultaneously.
3. Their signals interfere, causing a collision.
4. Both hosts detect the collision, send jam signals, and stop normal transmission.
5. Each chooses a random backoff interval.
6. One host retries first and completes its frame; the other waits and retries later.

One Endpoint per Switch Port

Each endpoint has a separate collision domain toward the switch. The host and switch can transmit simultaneously because the link is full duplex. No CSMA/CD collision handling is required.

Busy Legacy Hub Segment

As more hosts become active, more stations compete for the same medium. Contention, collisions, and retransmissions increase, reducing useful throughput. Replacing the hub with a switch and using full-duplex links removes the shared collision domain.

Duplex Mismatch

A duplex mismatch occurs when two connected interfaces use incompatible duplex modes. For example, one side may operate full duplex while the other operates half duplex.

The half-duplex side believes it must use CSMA/CD. It may report collisions or late collisions when the full-duplex side transmits without following shared-medium rules. The full-duplex side does not perform CSMA/CD, but it may receive damaged or overlapping frames and report CRC/FCS, alignment, or runt-related symptoms.

This asymmetric symptom pattern is an important troubleshooting clue: collision-related counters on one side and frame-integrity errors on the other often point to a duplex mismatch.

CCNA Troubleshooting Workflow

1. Confirm the topology. Determine whether the interface connects to a hub, a shared legacy segment, or a switch. Verify whether the link should be half or full duplex.
2. Inspect speed and duplex negotiation. Check the operational values at both ends. Auto-negotiation is generally preferred when both devices support it.
3. Review interface counters. Compare collisions, late collisions, excessive collisions, CRC/FCS errors, alignment errors, runts, and utilization.
4. Inspect the physical path. Check cable length, connectors, transceivers, intermediary equipment, and legacy design-rule violations.
5. Correct the configuration. Make both ends use compatible speed and duplex settings. Do not assume that setting one endpoint to full duplex makes a hub-based shared segment full duplex.
6. Retest. Clear or record counters as appropriate, generate controlled traffic, and verify that the relevant error counters stop increasing.

Cisco IOS Commands

Use these commands to inspect interface state, counters, speed, and duplex:

show interfaces <interface>
show interfaces <interface> status
show interfaces counters errors

Where manual configuration is necessary and supported, matching settings can be applied on both ends. Prefer auto-negotiation on both ends when supported:

configure terminal
interface <interface>
speed auto
duplex auto

If manual settings are required, configure the same compatible speed and duplex mode at both ends. Configuration syntax and supported values vary by interface. A full-duplex setting on an endpoint does not convert a hub or shared medium into a full-duplex network.

Interpreting Common Symptoms

Key Exam Points

• CSMA/CD expands to Carrier Sense Multiple Access with Collision Detection.
• It applies to shared, half-duplex Ethernet.
• Carrier sense means listening before transmitting.
• Multiple access means several stations share one medium.
• Stations monitor the medium while transmitting to detect collisions.
• After a collision, stations send a jam signal, discard the collided frames, and use randomized backoff.
• Binary exponential backoff expands the retry range after repeated collisions.
• The classic slot time is 512 bit times.
• The classic minimum Ethernet frame size is 64 bytes; padding is used when necessary.
• Ethernet permits a maximum of 16 transmission attempts before reporting excessive collisions.
• A normal collision occurs within the slot-time window; a late collision occurs after it.
• Full-duplex switched Ethernet does not use CSMA/CD.
• Collision counters are distinct from CRC/FCS errors.
• A duplex mismatch can produce late collisions on the half-duplex side and CRC/FCS or related errors on the full-duplex side.