A switch gives every device its own private, collision-free wire. It still has to flood a broadcast frame out every other port on the same VLAN — those are two completely different jobs, and only one of them is solved by switching.
"Domain" shows up twice in networking vocabulary, describing two boundaries that used to line up almost perfectly on old hub-based networks and now almost never do. A collision domain is the set of devices whose transmissions can physically collide with each other on a shared medium — a real electrical event that shared-media Ethernet (hubs, coax) had to detect and recover from. A broadcast domain is the set of devices that receive a copy of a broadcast frame — a Layer 2 delivery boundary, not a physical-collision boundary at all. A modern switch happens to eliminate the first kind of domain almost entirely, port by port. It does nothing whatsoever to shrink the second — and that gap is where a lot of real-world "why is this switched network still slow" problems come from. This explainer builds directly on the companion piece on switching vs. routing — that one covers what switches and routers do with a frame; this one covers the scope boundaries their forwarding behavior creates.
On old shared-media Ethernet — a hub, or coax cable with multiple taps — every connected device was electrically attached to the exact same physical transmission medium. There was no dedicated pair of wires between any two devices; there was one shared channel that all of them took turns using. If two devices happened to transmit at the same instant, their electrical signals physically overlapped on that shared medium and corrupted each other — a genuine collision, not a metaphor. Ethernet handled this with CSMA/CD (Carrier Sense Multiple Access with Collision Detection): a device listens before transmitting, and if a collision is detected anyway (because timing on a shared medium can never fully rule it out), both devices stop, wait a random backoff interval, and retransmit. Every device hanging off a hub belonged to the same single collision domain, because every device shared the same physical medium where a collision could occur.
A switch is a fundamentally different kind of device from a hub, even though both look like "a box with ports." Instead of repeating every signal onto one shared internal medium, a switch gives each connected device its own dedicated point-to-point connection, almost always running full-duplex — meaning the device can transmit and receive simultaneously without ever contending for the wire with anything else. With no shared medium and no other device electrically present on that same link, there is nothing left to collide with. Each switch port is its own separate collision domain — on a fully switched, full-duplex network, collisions in the classic CSMA/CD sense have effectively been engineered out of existence. That part of the old hub-era problem is genuinely, structurally solved.
A switch eliminates collision domains by giving each port its own dedicated, full-duplex connection — isolation is exactly the mechanism, and isolation is a good thing for collisions. But a broadcast frame's entire purpose is to reach every device in the domain, and a switch's Layer 2 forwarding behavior is specifically defined to honor that: flood it out every port except the one it arrived on, within the same VLAN. A switch that didn't do this would be broken — ARP, DHCP discovery, and plenty of other essential Layer 2 mechanisms depend on broadcasts actually reaching everyone. So switching structurally cannot shrink a broadcast domain, no matter how many ports or how much forwarding intelligence the switch has. The only thing that ever shrinks a broadcast domain is a Layer 3 boundary — a router interface, or a Layer-3-capable switch doing inter-VLAN routing — because routing operates per-subnet, and broadcasts are, by definition, never routed between different networks. That's exactly why VLANsexist: they're a deliberate way to carve one large switched Layer 2 network into several smaller broadcast domains, each small enough that broadcast traffic stays a non-issue.
Half right, and the wrong half matters. Switches genuinely do eliminate collision domains — each port is its own isolated, full-duplex connection, and classic Ethernet collisions are effectively gone on a fully switched network. But switches do not shrink broadcast domains — a switch is specifically required to forward a broadcast frame out every port within the same VLAN, which is the opposite of shrinking anything. A large, flat switched network — one giant VLAN spanning hundreds of devices — still has exactly one large broadcast domain, and it can absolutely suffer real performance problems from broadcast traffic: every ARP request, every DHCP discovery, every misbehaving broadcast-chatty application gets flooded to every single device, consuming bandwidth and CPU cycles on hosts that have nothing to do with it. Only VLAN segmentation — splitting that one large broadcast domain into several smaller ones, paired with the Layer 3 routing needed to let those VLANs talk to each other — actually addresses that. This is exactly why VLANs remain a standard, necessary design practice even on networks built entirely out of modern, fully switched hardware.
Explains the difference between a collision domain (a set of devices whose transmissions can physically collide on a shared medium, historically relevant to hub/coax Ethernet and resolved per-port by modern switches) and a broadcast domain (the set of devices that receive a broadcast frame, bounded only by Layer 3/VLAN boundaries and left completely unchanged by Layer 2 switching).
A collision domain is the set of devices that share one physical transmission medium closely enough that two simultaneous transmissions can physically collide and corrupt each other. This was a real, common event on hub-based and coax Ethernet, handled by CSMA/CD (carrier sense, collision detection, random backoff, and retransmission). A modern switch gives every connected device its own dedicated, full-duplex point-to-point link — with no shared medium and nothing else present on that link, a collision becomes structurally impossible. Each switch port is therefore its own separate collision domain, and fully switched, full-duplex networks have effectively eliminated collisions as a practical concern.
A broadcast domain is the set of devices that receive a copy of a frame addressed to the broadcast MAC address (FF:FF:FF:FF:FF:FF) — a frame meant for every device on the segment. A switch is specifically required to forward broadcast frames out every port except the one they arrived on, within the same VLAN; that's the correct, necessary behavior that lets protocols like ARP and DHCP discovery work at all. Because of this, any number of switches trunked together in the same VLAN still form one single, shared broadcast domain — switching intelligence and port count have no effect on broadcast domain size. Only a Layer 3 boundary — a router interface, or a Layer-3-capable switch performing inter-VLAN routing — actually stops a broadcast frame from propagating further, because routing happens per-subnet and broadcasts are never routed between different networks.
The two behaviors come from opposite design goals. Eliminating collision domains is about isolation: give each device its own private link so nothing else is ever present to collide with. Preserving (not shrinking) the broadcast domain is about reachability: a broadcast's entire purpose is to reach every device on the segment, and a switch's Layer 2 forwarding rules are built specifically to honor that by flooding it everywhere within the VLAN. A switch that suppressed broadcasts instead of flooding them would break basic network operation. This is precisely why an oversized, flat switched network — one giant VLAN spanning hundreds of devices — is a genuine, common scalability problem: switching alone cannot fix broadcast traffic overhead, no matter how fast or intelligent the switch is.
Because switching can't shrink a broadcast domain, network designers use VLANs specifically to do it deliberately — segmenting one large Layer 2 network into several smaller broadcast domains, each sized so broadcast traffic (ARP, DHCP, and similar chatter) stays a small fraction of overall bandwidth rather than a bottleneck. VLANs alone only solve the broadcast domain size problem, though — devices in different VLANs can no longer reach each other by switching, so a Layer 3 device (a router, or the same switch acting as a Layer 3 switch) is also required wherever those VLANs need to communicate. Segmenting broadcast domains and providing inter-VLAN reachability are a matched pair of design decisions, not one feature.
Effectively, yes, but they're trivial: each switch port with a single device attached is its own one-device collision domain, and because the link is full-duplex and point-to-point, no actual collision can ever occur on it. In practice, people describe fully switched, full-duplex networks as having eliminated collisions rather than having 'one collision domain per port that never collides,' though both descriptions are technically accurate.
Because broadcast domain size is defined by Layer 3/VLAN boundaries, not by switching capacity or intelligence. A switch's job when it receives a broadcast frame is to forward it out every other port in the same VLAN — that behavior doesn't change no matter how many switches are chained together via trunk links or how fast they operate. Ten fast switches trunked together in one VLAN are still one broadcast domain, just as much as a single slow one would be.
Every broadcast frame — ARP requests, DHCP discovery packets, and any broadcast-heavy application traffic — gets delivered to every device in the domain, consuming bandwidth and forcing every host's NIC and CPU to process it, even hosts with no interest in the traffic. In a large enough flat VLAN (hundreds of devices), this 'broadcast noise' can measurably degrade performance and, in severe cases (like a Layer 2 broadcast storm from a switching loop), can saturate links entirely.
No — a Layer-3-capable switch stops broadcasts at a VLAN/subnet boundary using the same principle as a dedicated router: it only forwards traffic between VLAN interfaces at Layer 3 (by IP address, via routing), and broadcasts are never part of that inter-VLAN routing decision. The 'router' function is what stops the broadcast, whether it lives in a standalone router chassis or inside a Layer 3 switch.
Not directly — collision domains are already handled per switch port by full-duplex, dedicated links, independent of VLAN assignment. VLANs exist purely to manage broadcast domain size (and related Layer 2 segmentation/security concerns), which is a separate axis from the collision-domain problem that switching already solved.
Try our Enterprise IT Networks Studio
More calculators, simulators, and guides for this discipline.