101
0/228
Loading content...
When two Ethernet stations transmit simultaneously and their signals collide, which other stations are affected? In a small network with a single hub, everyone. In a larger network with multiple segments, perhaps only a subset of stations. The collision domain is the set of all devices that will experience the collision—the arena within which contention for the shared medium occurs.
Understanding collision domains was critical for early network design and remains essential for understanding why modern switched networks perform so much better than their hub-based predecessors.
By the end of this page, you will understand how collision domains are defined, bounded, and segmented. You'll learn how different network devices affect collision domain scope, how to calculate the number of collision domains in a network, and why reducing collision domain size dramatically improves network performance.
A collision domain is a network segment where packet collisions can occur. More precisely, it's the set of all network devices that share the same medium and whose simultaneous transmissions would result in a collision that both (or all) parties would detect.
Think of it like an open office space with a single announcement system. If two people try to make announcements simultaneously, everyone hears the garbled result. The entire office is a single "announcement domain." Adding soundproof conference rooms creates separate domains where announcements don't interfere.
Collision domains exist only in half-duplex environments where devices share a transmission medium. Full-duplex point-to-point links (like modern switch-to-device connections) have no collision domain because each direction has its own dedicated channel.
Different network devices create different effects on collision domain boundaries. Understanding which devices extend versus terminate collision domains is fundamental to network design.
| Device | OSI Layer | Collision Domain Effect | Bandwidth Sharing |
|---|---|---|---|
| Coaxial Cable Segment | Layer 1 | All connected devices in same domain | Shared 10 Mbps |
| Repeater | Layer 1 | Extends/joins collision domains | Shared bandwidth |
| Hub | Layer 1 | All ports in same collision domain | Shared bandwidth |
| Bridge | Layer 2 | Separates collision domains | Dedicated per segment |
| Switch | Layer 2 | Each port is separate domain* | Dedicated per port |
| Router | Layer 3 | Separates collision domains | Dedicated per interface |
Layer 1 Devices: Repeaters and Hubs
Repeaters and hubs operate at the physical layer. They receive electrical signals on one port, regenerate them, and transmit them out all other ports. This is pure signal amplification with no interpretation of frame contents.
From a collision perspective, devices connected through repeaters or hubs behave as if they're all on the same physical cable. A transmission on any port is repeated to all ports, and a collision anywhere is experienced everywhere.
Layer 2 Devices: Bridges and Switches
Bridges and switches operate at the data link layer. They examine the destination MAC address in each frame and make forwarding decisions:
This intelligent forwarding terminates collision domains. Frames from one port don't automatically propagate to other ports unless necessary, preventing collisions from spanning across the entire network.
In a modern switched network with full-duplex links, each switch port effectively has its own collision domain with exactly one device. Since that single device uses full duplex, collisions can't occur at all. This is why switch-based networks have no practical collision concerns.
The size of a collision domain—measured both in number of devices and physical extent—has profound effects on network performance. Larger collision domains suffer from multiple compounding problems.
Problem 1: Increased Collision Probability
With more devices sharing the medium, the probability that two or more will attempt simultaneous transmission increases. For n devices, each with load λ, the collision probability grows approximately as:
$$P_{collision} \propto n^2 \times \lambda^2$$
Doubling the number of devices approximately quadruples the collision rate (all else being equal).
Problem 2: Reduced Per-Device Bandwidth
All devices share fixed total bandwidth. A 10 Mbps hub with 20 devices provides an average of 500 Kbps per device—and that's before accounting for collision overhead.
Effective bandwidth per device:
$$BW_{effective} = \frac{BW_{total} \times Efficiency}{n}$$
As n increases and efficiency decreases (more collisions), effective per-device bandwidth plummets.
| Stations | Collision Probability* | Utilization Cap** | Effective BW/Device*** |
|---|---|---|---|
| 2 | ~1% | ~90% | ~4.5 Mbps |
| 10 | ~10% | ~70% | ~700 Kbps |
| 25 | ~25% | ~50% | ~200 Kbps |
| 50 | ~40% | ~35% | ~70 Kbps |
| 100 | ~50% | ~20% | ~20 Kbps |
Problem 3: Longer Propagation Delays
Larger collision domains typically span greater physical distances. This increases propagation delay, which:
Problem 4: Cascade Collisions
When many stations back off from a collision, they often choose similar backoff intervals (especially in early retransmission attempts with small backoff windows). This leads to cascade collisions where the same stations collide repeatedly.
In large collision domains with many waiting stations, the network can enter a state of collision collapse where almost all transmission attempts result in collisions, and throughput drops to near zero despite high demand.
The Solution: Segmentation
The cure for large collision domain problems is segmentation: dividing one large domain into multiple smaller ones using bridges, switches, or routers. Each smaller domain has fewer devices, shorter propagation delays, and independent bandwidth.
A common exam and interview question asks: "How many collision domains are in this network?" Let's develop a systematic approach to answering this question.
Example Network Analysis:
Consider a network with:
Collision Domain Count:
Router interfaces: 3 collision domains (including the switch connection and 2 other networks)
Switch direct connections: Each of the 20 PCs directly connected to the switch has its own collision domain = 20 collision domains
Switch to router: Counted as part of router's domain (1 domain but already counted)
Switch to Hub 1: All 8 PCs through Hub 1 share 1 collision domain
Switch to Hub 2: All 7 PCs through Hub 2 share 1 collision domain
Total: 3 (router) + 20 (direct switch) + 1 (hub 1 segment) + 1 (hub 2 segment) = 25 collision domains
Note: The switch ports connecting to hubs become part of those hubs' collision domains, so we don't count them separately.
Don't count the number of devices—count the domains! A hub with 100 devices is still just 1 collision domain. A switch with 100 devices is 100 collision domains (one per port, assuming one device per port).
Students often confuse collision domains with broadcast domains. While related, they represent different network properties and are bounded by different devices.
Key Relationship:
Example:
A network with:
Collision domains: 2 (router interfaces) + 24 + 24 = 50 (each switch port plus router interfaces)
Broadcast domains: 2 (one per router interface/subnet)
All 24 PCs behind a switch share a broadcast domain—a broadcast from any of them reaches all 23 others (and the router). But each PC has its own collision domain through the switch—transmissions from one don't collide with transmissions from others.
| Device | Collision Domains | Broadcast Domains |
|---|---|---|
| Hub | Extends (single domain) | Extends (single domain) |
| Bridge | Separates | Extends |
| Switch | Separates (per port) | Extends |
| Router | Separates | Separates |
| Layer 3 Switch | Separates (per port) | Separates (per VLAN) |
In classic shared Ethernet, the collision domain's physical extent was strictly limited to ensure CSMA/CD timing requirements were met. These limits were codified in the famous 5-4-3 rule.
The 5-4-3 Rule for 10BASE5 and 10BASE2:
Within a single collision domain:
This rule ensures that the worst-case propagation delay stays within the 512 bit-time slot time, enabling reliable collision detection.
Why This Configuration?
Each component adds delay:
With maximum segments (500m × 5 = 2500m) and 4 repeaters:
The 5-4-3 rule is historically important but irrelevant to modern networks. Today's switched networks don't share collision domains across segments, so the timing constraints don't apply. However, similar calculations still matter for full-duplex link distance limits based on signal attenuation and link timing.
| Standard | Max Segment | Max Segments | Max Diameter | Max Repeaters |
|---|---|---|---|---|
| 10BASE5 | 500 m | 5 | 2500 m | 4 |
| 10BASE2 | 185 m | 5 | 925 m | 4 |
| 10BASE-T | 100 m | Depends on hubs | ~500 m | 4 (Class I) or 2 (Class II) |
| 100BASE-TX | 100 m | Limited | ~200 m | 1 (Class I) or 2 (Class II) |
When collision domain size becomes a performance problem, network designers employ segmentation strategies to divide the domain and improve throughput.
The Switch Revolution:
The transition from hubs to switches was one of the most significant upgrades in network history:
Before (Hub-based):
After (Switch-based):
This is a 25× improvement in per-device bandwidth with no change to the wiring or end devices—just replacing the hub with a switch.
When Full Duplex?
Full-duplex operation requires:
In full duplex, collision domains don't exist because there's no shared medium. The "collision domain" concept is simply not applicable.
A common configuration error is duplex mismatch—one end in full duplex, the other in half duplex. The full-duplex end transmits whenever it wants (it doesn't sense carriers or detect collisions). The half-duplex end sees constant collisions and backs off repeatedly. This causes severe performance degradation and is a common troubleshooting target.
Collision domains are foundational to understanding Ethernet network design and the evolution from shared to switched networks. Let's consolidate the key concepts:
What's Next:
Now that we understand collision domains and their boundaries, we need to examine the timing that makes CSMA/CD work within a collision domain. The slot time is the fundamental timing unit that drives collision detection, backoff calculations, and minimum frame size requirements. We'll explore slot time in depth in the next page.
You now understand collision domains—how they're defined, bounded, calculated, and how they differ from broadcast domains. This knowledge is essential for network design, troubleshooting, and understanding the evolution from hub-based to switch-based Ethernet networks.