Connecting Multiple Devices in Ethernet
Two computers can communicate through one network cable. This is a point-to-point connection. With three, four, or ten computers, direct connections between every pair require many cables and separate interfaces, so the arrangement quickly becomes impractical. The number of links grows with every additional pair of devices, which is why local networks use a shared connection point.
Central networking devices provided that connection point. Each computer received its own cable to the shared center. During Ethernet's transition from bus topology to star topology, the device used for this purpose was the hub, also known as a concentrator. It allowed a local area network (LAN) to use simple cabling. The physical cables were separated, while transmission still used one shared channel. The arrangement reduced the number of long cable segments, while it preserved the contention rules of shared Ethernet.
An Ethernet hub as a Layer 1 repeater
A hub is a multiport signal repeater that operates at Layer 1, the Physical Layer, of the . This layer defines the hub's function. The device processes electrical signals and does not analyze frame headers.
A hub does not make forwarding decisions from the contents of transmitted data. It does not recognize:
- Frames or packets: It does not analyze header fields or the type of the transmitted unit.
- MAC addresses: It does not read source and destination MAC addresses to filter traffic.
- IP addresses: It does not interpret network-layer addresses such as IP.
A hub acts as a multiport repeater. It receives an electrical signal representing bits on one port, restores its level, and repeats it through the other ports. Every connected interface therefore receives signals from the other sections of the shared medium. The hub does not store a frame, select a port from a destination address, or create separate queues for different connections. It also does not divide the collision domain between ports. Those properties distinguish it from a switch, which receives a complete frame and consults a forwarding table for forwarding decisions.
Physical star and logical bus
Hub-based networks used a physical star topology. Each device had its own cable to the central hub. This changed the cabling used by the older bus topology, as in 10BASE-2, where devices shared one cable. A break in a cable to the central device disconnected one station, which made locating that fault easier.
The physical arrangement resembled a star, while the hub created a logical bus topology. It repeated every signal through the other ports, so all interfaces shared one communication medium. In half-duplex mode only one station could transmit at a time. This shared medium formed one collision domain covering the hub ports. The central cabling simplified individual fault isolation, while the access rules of the shared channel remained unchanged. Every port took part in the same transmission contention and received the signals present on the segment.
One collision domain
A is a network segment where devices share a medium and can begin transmitting at the same time. Their signals then overlap, and the stations must retransmit. A hub creates one collision domain that includes all of its ports. All devices connected to the shared medium belong to the same domain.
Consequences of a Single Collision Domain
- Half-duplex operation:The shared electrical path permits transmission in one direction at a time. Communication uses . A device listens before transmitting and does not transmit in both directions at once.
- Lower throughput: Adding devices to a hub or cascading hubs increases the number of stations competing for the same medium. After a collision, stations stop transmitting, choose a random backoff time, and retry. The waiting period and retransmissions consume part of the available capacity, so traffic from many stations lowers the throughput observed by each station.
- CSMA/CD: A shared medium requires the access rules defined by CSMA/CD. An interface listens to the channel, starts transmission after detecting an idle medium, and checks for a collision. After a collision it sends a jam signal, chooses a backoff interval, and retries the frame. The hub repeats signals and propagates the collision, while half-duplex Ethernet interfaces perform this procedure.
Hub throughput and traffic confidentiality
Repeating a signal through every other port affects medium utilization and allows other stations to receive the same traffic.
Shared bandwidth
In a 10-port hub operating in a 100 Mbps network, the shared medium provides 100 Mbps for the entire collision domain. That capacity is shared by all ports. If Host A sends a file to Host B while Host C sends a file to Host D, both transmissions compete for the same capacity. The hub repeats the signals carrying each frame through the other ports, so stations outside a given flow also receive the signal and occupy their links. Port rates do not add together for parallel conversations because all transmissions use the same medium limit.
No traffic isolation
A hub does not filter traffic by destination address. Signals carrying each frame are repeated through the other ports, so an interface connected to the hub can receive frames intended for other stations. A user can place the network card into and use a packet analyzer such as Wireshark to record traffic visible on the segment. Confidentiality then depends on encryption at higher layers because the hub provides no isolation between ports.
Replacing hubs with switches
Hubs simplified the construction of star-topology networks. Cabling from stations to a central device was easier to service than the coaxial bus systems used earlier.
The limits of shared media led to the adoption of the Ethernet switch. A switch operates at Layer 2, learns source MAC addresses, and records their association with ports and VLANs. For a known destination address it forwards the frame to the corresponding port. Broadcast frames and unknown unicast destinations are flooded within the relevant VLAN. Two full-duplex links can carry transmissions simultaneously when the devices and their configuration support that mode.
Each switch port forms a separate collision domain, and a full-duplex link does not use CSMA/CD. Link capacity is not shared with the other ports, while the switch still forwards broadcasts and flooded frames within a VLAN. Hubs are now a historical technology, useful mainly for explaining shared Ethernet. Replacing a hub with a switch does not by itself describe the whole forwarding model, which also includes the MAC table, broadcast domains, and VLAN configuration. Collision-domain scope and VLAN scope should be assessed separately when estimating visibility and capacity.