A star topology connects every node to one central hub, normally a switch, each with its own cable. No two nodes connect directly. A fault affects one node and the port names it, which is the advantage that ended bus and ring networks on the LAN.
- One cable per node, and no node depends on another
- A fault affects one node, and the switch port identifies it
- The central hub is the single point of failure, and that is the trade
- Copper reaches 100 meters, which is a radius, not a length
- Real buildings run an extended star: switches uplinked to switches
On this page
- How a star topology moves data
- Passive and active stars
- Why the star topology beat the alternatives
- What the center gives you: management and performance
- The central hub as a single point of failure
- The extended star topology, which is what buildings run
- The 100 meter radius nobody plans for
- Where else a star topology shows up
- Where people go wrong
- Comparison
- FAQ
How it worksHow a star topology moves data
A star topology has two kinds of member. One central node sits in the middle, and every other connected device is a peripheral node with a single cable to it. No peripheral device is wired to another peripheral device, and that fact produces everything else about the star topology.
Data leaves a device, arrives at the central hub, and the hub decides where the data goes. On a modern switch that decision comes from the MAC address table, so the data leaves by one port and no other device on the network sees it.
On a repeater hub, which is what the center of a star topology used to be, the data went out every port and every computer saw everything, which is why repeater hubs are extinct.
Transmission in a star topology is therefore always indirect. A peripheral node hands its data to the central node, which forwards it toward the destination node, so every exchange between two connected devices passes through the middle.
A switch makes that indirection cheap. Each cable is a collision domain of its own, so transmission on one link never collides with transmission on another.
Two consequences follow immediately, and they are what made star networks easy to run.
All communication is two hops. Device to central hub, hub to device. That is the whole path between two connected devices across a single switch, and it is why latency inside a star topology is the switch forwarding delay and nothing else.
No device depends on any other device. A computer that is powered off, unplugged or broken is invisible to the rest of the network. On a shared bus or an unprotected ring topology the same failure takes out its neighbors, and that difference is the practical argument that ended both of those networks.
Passive and activePassive and active stars
Two types of star topology appear in older material, and the distinction still explains the vocabulary.
A passive star topology has a central device that only joins the cables and passes signals through without regenerating them. Optical splitters in a passive optical network work this way, and so did some early wiring systems. Nothing amplifies, so the distance budget is spent on the whole path rather than per segment.
An active star topology has a central hub that regenerates every signal before sending it on. Every Ethernet star network is active, whether the middle is an old repeater hub or a modern switch, and it is why the 100 meter limit applies from each device to the hub rather than across the network as a whole.
The practical value of knowing this is small until somebody hands you a passive optical system and expects it to behave like an Ethernet switch. The two star topologies look identical on a diagram and their distance rules are completely different.
Why it wonWhy the star topology beat the alternatives
Three arguments decided it against the other network topologies, and only one is about networking at all.
The building is already a star. Structured cabling standards specify a home run from every outlet back to a telecommunications room. Contractors quote it that way, floor plans assume it, and the patch panel in the closet is the central hub of a star topology whether or not anybody used the words.
A network topology that matches the cables already in the walls has an insurmountable advantage over any topology that does not.
Moves, adds and changes are easy. Adding a device to the network is one cable connection into a free port on the hub. Removing one is unplugging it. Nothing else on the network is touched or interrupted.
In a ring topology the loop has to be broken to insert a device, and on a bus the whole segment is interrupted. In an office where people change desks constantly, this is the argument that actually wins.
Faults are easy to isolate and already named. One bad cable affects one computer, and the port counters on the central hub say which port and what kind of fault. Working out which segment of a shared medium is faulty means bisecting the whole cable plant. That difference is measured in hours per incident.
The cost of all this is cables. A star topology needs one run per device where a bus network needed one shared run, and in a building where the trench or the ceiling void is the expensive part, that mattered enormously in 1990 and matters very little now that the labor is the cost and the cable is not.
ManagementWhat the center gives you: management and performance
Two advantages of the star topology follow from having one device in the middle, and neither is obvious from a diagram.
Network management has one place to happen. Every device is on its own port, so VLANs, access control, port security, quality of service and power over Ethernet are all configured per device from one console.
Monitoring works the same way: the hub counts bytes, errors and discards per port, and mirroring a single port copies exactly one computer's data to an analyzer. On a shared medium none of that is possible, because there are no per-node ports to configure or measure.
Security in a star network is a per port question for the same reason. Access control can be applied to one device without touching any other, which no shared medium allows.
Performance is per node rather than shared. Each cable to the central hub is a link of its own, so two computers exchanging data at gigabit speed do not slow down a third.
Full duplex on every port removes collisions entirely, which is what a repeater hub could never do, and the switch backplane is normally fast enough that all ports can run at once. The performance ceiling of a star network is the uplink out of it, not the topology.
Those two properties are why the star topology survived the arrival of everything that came after it. VLANs, PoE phones, wireless access points and network access control all assume one device per port, and only this topology provides that.
The single pointThe central hub as a single point of failure
The central hub takes the whole network with it when it fails, and this is the one real objection to the star topology.
It is worth being precise about how bad that is. A switch is a solid-state device in a rack with no moving parts except fans, and it fails far less often than the cables and the computers around it. The realistic failure is a power loss or a configuration mistake, not the hardware.
Three mitigations exist for star networks, and they escalate in cost.
Uninterruptible power. Most switch outages are power outages rather than equipment failures. This is the cheapest fix and the one most often missing.
A second hub with devices split across both. Half the users lose service instead of all of them, which turns a network outage into a degradation. This costs one extra switch and some patching discipline.
Stacked or chassis switches with redundant supplies and uplinks. Two devices behaving as one central hub, so a failure of either leaves the network up. This is the standard answer once downtime has a number attached to it.
What none of these change is that a device with one cable to one hub depends on that hub. Dual-homing a workstation is not done and is not worth doing. Dual-homing a server is, and that is where the redundancy budget belongs.
Extended starThe extended star topology, which is what buildings run
A single central hub covers a small office network. Anything larger is a hierarchy of star topologies, one per switch, all the way up.
Access switches live in the closet on each floor, each the central hub of a star of user cables.
Distribution switches sit in the main room, with a star of uplink cables from the access switches around them.
The core, in large enough networks, is the star those distribution switches point at.
That structure is called an extended star topology, or a tree topology, and it is not a different design so much as the star topology repeated.
It has one property worth understanding: the uplink between two switches carries all the data of every device behind it, so an access switch with 48 devices on gigabit ports and one gigabit uplink is oversubscribed 48 to 1. That number is almost never a problem and is always worth knowing.
Redundant uplinks introduce a loop, and a loop in an Ethernet network is a broadcast storm, which is why spanning tree exists and why it runs on every switch in every building whether or not anybody configured it.
The 100 m radiusThe 100 meter radius nobody plans for
Copper Ethernet is limited to 100 meters between a device and the central hub, and the standard splits that as 90 meters of fixed cables plus 10 meters of patch leads at both ends.
That limit is the real constraint on how many closets a building needs. It is a radius from the central hub, not a cable length, so floors bigger than roughly 90 meters in any direction from the closet need a second closet or a fiber run to an intermediate switch.
This is the one place where the physical layout of a star topology drives the budget rather than following it.
Two things go wrong with it in practice. Long runs installed at 95 meters work on the day and fail later as connectors age or the link renegotiates to a lower speed.
And the 10 meter patch allowance is spent quickly by a long lead at the desk and a badly managed patch panel, so a 90 meter permanent link plus a 15 meter patch lead is out of specification even though nothing about it looks unusual.
Other starsWhere else a star topology shows up
Office cabling is the obvious case. The same shape runs several systems that nobody describes as networks at all, and recognizing it saves time when one of them breaks.
Home networks. Every connected device in a house meets at the internet box, which holds the switch and the access point. That box is the central node of a star topology whatever the cables look like.
Wireless. Clients associate with an access point and send everything through it, including data bound for the laptop at the next desk. A wireless network is a logical star whose transmission medium is shared, so the per link independence of a cabled star is the part it does not get.
Passive optical networks. One fiber feed splits to many premises from a single point, which is a passive star topology. The split is shared capacity rather than a port per device.
Building and control systems. Access control panels, camera systems and industrial controllers are wired as home runs back to one controller, for the same fault isolation reasons that decided it for data networks. These types of star topology share the shape and not the medium, and the medium is what decides how they behave under load.
PitfallsWhere people go wrong
Treating the central hub as the main risk. It fails less often than everything around it on the network. Power protection and a second switch for critical users cost less than the redundancy people plan instead.
Cabling two switches together twice without spanning tree. The redundancy you wanted becomes a broadcast storm within seconds. The protection has to be running before the second cable goes in.
Ignoring uplink oversubscription until a backup job runs. Forty-eight gigabit computers behind one gigabit uplink is fine all day and not fine at 2 a.m. Size the uplink for the data that crosses it, because it is the only real performance limit in a star network.
Running copper past 100 meters because it worked when tested. It is a specification limit, not a suggestion, and the failures it produces appear months later and look like something else entirely.
Assuming a star topology means one switch. Real networks are extended stars, and the interesting questions are about the uplink cables between the tiers rather than about the edge ports.
Buying a bigger switch instead of a second one. A larger central hub concentrates more nodes in one box. Two smaller switches with devices split across them is usually the better spend at the same price.
ComparisonFour topologies, and the two rows that point in opposite directions
| Criterion | Star | Ring | Bus | Full mesh |
|---|---|---|---|---|
| Cables for n nodes | n | n | One shared | n(n-1)/2 |
| Adding a node | One free port, easy | Breaks the loop | Interrupts the segment | A link to everyone |
| A single cable fault | Affects one node | Splits the network | Kills the segment | Affects one link |
| Fault isolation | Per port, named | Around the loop | Very poor | Per link |
| Single point of failure | The central hub | An unprotected break | Any break | None |
| Latency between nodes | Two hops | Grows with node count | Shared | One hop |
| Used today | Every office network | Metro and industrial | Nowhere | Data center fabrics |
The row that decides it inside a building is adding a device, and the row that decides it for a fiber path outdoors is the cable count. Those two rows point in opposite directions, which is exactly why one network topology owns the inside of buildings and another owns the outside.
FAQFrequently asked questions
What is a star topology?
A network topology where every node connects to one central hub, normally a switch, with its own cable, and no two nodes are connected directly.
How does data travel in a star network?
From the device to the central hub, which looks up the destination and forwards the data out one port. All communication is two hops.
What are the advantages of a star topology?
A fault affects one node rather than the whole network, the hub port names which one, adding or removing devices is easy, network management and monitoring happen per port in one place, and the layout matches how buildings are already cabled.
What are the disadvantages of a star topology?
The central hub is a single point of failure, the network needs one cable per device, and the performance of that hub limits every device connected to it.
What happens if the central hub fails?
Every device connected to it loses service. This is the accepted trade, mitigated with uninterruptible power, a second switch, or stacked switches with redundant supplies.
Is a star topology the same as a hub and spoke?
The shape is the same at a different scale. Star usually describes devices around a switch inside a building, and hub and spoke describes sites around a central site on a WAN.
What is an extended star topology?
Switches uplinked to other switches, so the network is a hierarchy of stars. This is what almost every real building runs, and it is sometimes called a tree topology.
How far can a device be from the switch?
One hundred meters over copper, made up of 90 meters of fixed cabling and 10 meters of patch leads. Beyond that the link needs fiber or an intermediate switch.
Why did the star topology replace bus and ring networks on the LAN?
Because a fault affects one node instead of the whole network, adding a node is one easy port change, and buildings were already cabled as home runs to a closet.
Does a star topology need a switch, or will a repeater hub do?
A repeater hub works and copies all data to every port, which wastes bandwidth and shows every computer everyone else's traffic. Switches replaced hubs entirely and there is no reason to use one now.
How many nodes can one star topology support?
As many as the central hub has ports, then you uplink another switch and the network becomes an extended star topology. The practical performance limit is the uplink, not the port count.
Why is a star topology easier to manage?
Because every node has its own port on one device. VLANs, port security, quality of service, power over Ethernet and per-port monitoring are all configured and measured from a single console, which no shared medium allows.
What is uplink oversubscription?
The ratio between the total speed of the edge ports and the speed of the uplink carrying their traffic. Forty-eight gigabit ports behind one gigabit uplink is 48 to 1, which is normal and worth knowing.
Are wireless networks star topologies?
Logically yes. Every client associates with an access point and communicates through it rather than with other clients directly, which is the same shape without the cables.
What is the difference between a passive and an active star?
A passive star joins the cables and passes signals through without regenerating them, as an optical splitter does. An active star regenerates every signal at the central hub, which is what every Ethernet network does and why the distance limit is measured per cable rather than across the whole system.
Keep readingRelated concepts
Read next · Design Ring Topology, and the Networks Where It Never Went Away The topology this one beat indoors, and the one that beat it everywhere outside a building. Open this next11 min- Switching · 11 min Spanning Tree Protocol Two uplinks between the tiers is a loop, and a loop in Ethernet is a broadcast storm within seconds.
- Switching · 9 min The MAC Address Table, and How to Find Which Port a Device Is On The reason a switch in the middle beats a hub in the middle: it forwards out one port instead of all of them.
- Fundamentals · 10 min What a Network Hub Is, and Why the Switch Replaced It Why the repeater once at the center of the star became obsolete, and where one still hides.
- Fundamentals · 11 min What a Node Is, and Why the Word Changes Meaning The shape the nodes are arranged in.
- Design · 9 min Hybrid Topology, and Why You Already Have One The layer you actually design.