A network hub is a simple networking device that connects multiple Ethernet devices by repeating incoming electrical or optical signals out through other ports rather than selectively forwarding frames according to destination addresses. It operates at the physical layer, commonly described as Layer 1 of the OSI model, and does not maintain a MAC-address forwarding table like an Ethernet switch.
When a signal arrives on one port, the hub repeats it to its other relevant ports so that all connected devices on the shared medium can receive the transmission. Each device then determines whether the Ethernet frame is addressed to it. Because all devices share the same collision domain, simultaneous transmissions can interfere with one another. Traditional Ethernet hubs therefore used contention mechanisms such as Carrier Sense Multiple Access with Collision Detection. Hubs were historically common in local Ethernet networks, but they have largely been replaced by switches because switches provide substantially better performance, isolation and scalability while remaining relatively inexpensive.
The hub's operation is fundamentally different from that of a switch. A switch examines Ethernet frames and learns source MAC addresses so it can usually send a unicast frame only through the port leading to the destination device. A hub does not make that decision. It simply repeats the signal. If a computer connected to one hub port sends an Ethernet frame to another computer, the hub repeats the signal to the other ports.
Every connected device therefore receives the physical transmission, although only the intended recipient normally processes the frame as its own. This means that adding more devices to a hub does not create separate collision domains for each port. Instead, all devices connected to the hub participate in the same shared medium. The simplicity of this architecture was useful in early Ethernet networks, but its limitations became increasingly significant as network speeds and traffic volumes increased.
Hubs contributed to Ethernet's shared-medium architecture. When two devices transmitted at the same time, their signals could collide. The transmitting devices detected the collision and waited for a calculated interval before attempting retransmission. This mechanism was associated with half-duplex Ethernet and CSMA/CD. As Ethernet evolved towards switched, full-duplex operation, collision detection ceased to be a practical feature of ordinary modern Ethernet links because each switched port generally has a dedicated point-to-point connection and can transmit and receive simultaneously.
A hub cannot provide this same isolation because multiple devices share the same collision domain. Consequently, hub-based networks become increasingly inefficient as traffic increases. Collisions consume bandwidth and retransmissions add delay. Switches eliminate collisions between independent full-duplex ports, which is a major reason they replaced hubs in mainstream Ethernet installations.
Hubs also have bandwidth-sharing characteristics. If a hub supports a particular Ethernet speed, the connected devices share the capacity of the common collision domain rather than each receiving an independent full-duplex link. For example, a traditional 100 Mbps hub network does not provide every connected device with an independent 100 Mbps collision-free path to every other device. All participants contend for the shared medium.
By contrast, a modern 100 Mbps switch can provide separate full-duplex 100 Mbps links to each compatible endpoint, subject to the switch's internal forwarding capacity and uplink limitations. This difference becomes especially important when multiple devices communicate simultaneously. A switch can receive frames on several ports and forward them independently when the destinations are different, whereas a hub repeats signals across the shared environment. The switch therefore makes much more efficient use of available Ethernet capacity.
Hubs have some characteristics that can be useful for understanding network behaviour, even though they are rarely appropriate for modern production networks. Because a hub repeats signals to all ports, a network-analysis device connected to an appropriate port can potentially observe traffic from other devices on the same collision domain, subject to the capabilities of the physical interface and network conditions. This made hubs historically useful for certain troubleshooting and monitoring scenarios.
Modern switched networks require techniques such as port mirroring, network taps or specialised monitoring infrastructure to obtain equivalent visibility into traffic that would otherwise be sent only to its intended switch port. The fact that hubs expose shared traffic is also one reason they should not be considered a security improvement. Any confidentiality must come from higher-layer mechanisms such as encryption rather than from the hub itself.
Hubs can be classified according to how they handle signals, with passive and active concepts appearing in older networking terminology. A passive hub essentially provided a physical connection point without significant signal regeneration, while an active hub regenerated or repeated signals to extend the network within the limits of the relevant Ethernet standard. Intelligent hubs sometimes offered basic management information, but they still differed fundamentally from switches because they did not selectively forward frames based on learned MAC addresses.
The distinction between a hub and a switch is therefore not simply that one is “older” and the other “newer”; their forwarding architectures are different. A hub operates at the physical signal level, while a switch interprets Ethernet frames and makes forwarding decisions at the data-link layer.
Hubs became obsolete for most ordinary Ethernet deployments as switched Ethernet became inexpensive and widely available. Switches provide better bandwidth utilisation, separate collision domains, full-duplex operation and selective frame forwarding. They can also offer VLANs, Power over Ethernet, traffic monitoring, link aggregation, redundancy and other functions that hubs generally cannot provide. Even simple unmanaged switches are usually preferable to hubs for contemporary networks because they require little configuration while still providing the fundamental benefits of switched Ethernet.
Hubs may still appear in educational material, specialised laboratory environments, historical equipment collections or unusual diagnostic setups, but they are not a normal choice for connecting modern computers to an Ethernet network. Modern Ethernet standards and equipment are overwhelmingly designed around point-to-point switched links rather than shared hub-based segments.
The most important distinction is therefore that a hub repeats physical signals, while a switch forwards Ethernet frames selectively. A hub has no meaningful understanding of which MAC address belongs to which port, so it sends received signals to multiple ports. A switch learns source addresses, associates them with ports and uses those associations to make forwarding decisions. This difference affects collisions, bandwidth, privacy, performance and scalability.
A hub creates one shared collision domain, whereas each full-duplex switch port normally constitutes its own collision domain. A hub also cannot route between IP networks, just as a basic switch cannot normally route between IP subnets. Routing requires a Layer 3 function supplied by a router or Layer 3 switch. Understanding hubs is nevertheless important because they illustrate the historical shared-medium model from which modern switched Ethernet developed and make the architectural advantages of contemporary switching easier to understand.
