A mesh topology connects nodes directly to multiple other nodes, so data has more than one path between any two points. In a full mesh every node connects to every other node; in a partial mesh only the important ones do.
The picture in the textbook shows this as cabling between computers, and that network has never existed, because a computer has one network port and a full mesh of ten of them would need nine ports in each. Every mesh anybody actually runs is logical rather than physical.
- Every pair of nodes gets multiple possible paths for data
- A full mesh needs n(n-1)/2 links and n-1 interfaces on every node
- The wired diagram in the textbook is not a network anyone builds
- Real mesh networks are logical: routing tables, tunnels, or radio
- The three real ones: the internet core, WAN overlays, wireless
On this page
The arithmeticWhy the wired mesh in the textbook does not exist
Look at the standard mesh topology diagram: five or six computers with cables running between every pair. Then count the cables arriving at any one of those devices.
In a full mesh topology, each node needs an interface for every other node, so n-1 of them. Six computers means five network ports in every computer. Ten means nine. A PC ships with one, so the picture describes a device nobody sells for a network nobody has built.
| Nodes | Links | Interfaces on each node |
|---|---|---|
| 4 | 6 | 3 |
| 6 | 15 | 5 |
| 10 | 45 | 9 |
| 20 | 190 | 19 |
| 50 | 1,225 | 49 |
Both columns grow faster than the number of devices, and the right hand one is the one that kills the idea. Links are money. Interfaces are hardware that does not exist on an ordinary device.
This is not a small caveat about an otherwise sound design. It means the mesh topology as drawn is a teaching diagram for a property, which is fault tolerance through redundant paths, rather than a description of any network layout. The property is real and important. The cabling is not.
Full and partialFull mesh and partial mesh
The distinction is worth keeping because the second one is common and the first one is rare.
A full mesh topology connects every node to every other. Its single virtue is fault tolerance: no failure of one node or link isolates anything, because a path always remains.
Its cost is the table above, which is why in practice it appears only where the node count is small and the traffic is valuable: a handful of data centers, a few core routers, the sites of a business that cannot tolerate traffic taking the long way.
A partial mesh topology connects the devices that matter to multiple neighbors and leaves the rest single homed. This is what almost every resilient network is. The hub and spoke page works through the wide area version of that decision, where the arithmetic decides how much mesh you can afford.
The useful way to hold the two is as a dial rather than as two separate topologies. Every network sits somewhere between one link per device and every possible link, and moving along that dial buys fault tolerance with money.
How it worksHow a mesh network works
In a mesh network topology there is no central switch or hub that all communication passes through. Nodes are connected to several neighbors, and each node both sends its own data and forwards data for other nodes. Two techniques move a message across those connections.
Routing. The message travels along one path, hop by hop, chosen by routing protocols that know which connections are up. When a link or a node fails, the protocols calculate a new path around it. This is what vendors call self healing, and it is how the internet, wide area networks and most wireless mesh networks operate.
Flooding. Every node repeats the message to all of its neighbors until it reaches the destination. Flooding needs no routing tables and wastes bandwidth, so it suits small, low power networks. Bluetooth mesh uses a managed form of it.
During normal operation, traffic takes the shortest or cheapest path through the network. When something fails, traffic is rerouted through another node, and users notice nothing if the protocols converge quickly.
The same idea runs at very different scales. Zigbee, Thread and Z-Wave build mesh networks of sensors, locks and lights for smart home and IoT devices. 802.11s does it for Wi-Fi access points. OSPF and BGP do it for routers, across campus networks and the internet.
Pros and consMesh topology advantages and disadvantages
Compared with other network topologies, such as star, bus and ring, the mesh topology trades simplicity for redundancy. The advantages:
- Fault tolerance. No single point of failure, because data can be rerouted through other connections.
- Reliability under load. Several nodes can communicate at once without sharing one central link, so heavy traffic between two nodes does not slow the rest.
- Easy fault isolation. A failed link affects only the two nodes connected to it, which makes the fault quick to find.
- Coverage. In wireless networks, every added node extends the range of the whole mesh network.
The disadvantages:
- Cost. More links, ports and cabling than any other topology, growing much faster than the node count.
- Complexity. Installation, configuration and troubleshooting all take more work as connections multiply.
- Idle redundancy. Many redundant connections carry little or no traffic most of the time.
- Latency and power. In wireless and IoT networks, each hop adds delay, and nodes that relay for others use more power.
Where it is realThe three places a mesh actually exists
| Where | What is meshed | What the physical layer is |
|---|---|---|
| The internet core | Routing tables between autonomous systems | Fiber, mostly point to point |
| A wide area overlay | Tunnels between sites | Circuits to one provider each |
| Wireless mesh | Radio links between access points | Air, plus a wired uplink somewhere |
Read the third column. In none of the three networks is the mesh a set of cables between every pair of devices.
The internet core is the largest mesh anybody has built, and it is made of routing decisions. BGP gives each network several ways to reach a destination and a policy for choosing among them, which is exactly the property the textbook diagram is trying to teach, running on fiber that is nothing like meshed.
A wide area overlay builds a mesh out of tunnels. Each site has a circuit to a provider, which is a single link, and software builds direct encrypted paths between sites over the top. The mesh is in the tunnel table, and it can be full one week and partial the next without anybody touching a cable.
Wireless mesh is the one most people have met, in a house with several access points that relay for each other. IEEE 802.11s is the standard for it, issued in 2011 and folded into the base 802.11 standard in 2012, and it specifies a routing protocol so that units can find multi hop paths between themselves.
Here the mesh is genuinely between the devices, which is why it is the closest thing to the textbook picture that exists, and it is made of radio rather than cable.
The patternWhy every real mesh is logical
The pattern in all three is the same and it is the thing to take away.
A mesh is a property of the paths, not of the wiring. Give a set of devices multiple routes to each other and you have the fault tolerance the topology is drawn for, no matter what the cabling underneath looks like.
This is why a mesh can be reconfigured in software, why it can be partial in one direction and full in another, and why the answer to how many links your mesh has usually depends on which layer somebody is asking about.
It is also why mesh keeps returning under new names. A service mesh, an SD-WAN fabric and a wireless mesh have almost nothing in common physically. All three are the same idea: many paths, chosen dynamically, over an underlay that has far fewer.
Which raises the question the link count never answers. A second path helps only if something notices the first one is gone and moves traffic onto it, so the number worth asking about is not how many links the mesh has but how long it takes to converge after one fails.
That is a property of the routing protocol rather than of the topology: a wireless mesh reroutes in seconds, a tunnel overlay watching its own probes in under a second, and a network waiting on a routing timer can take minutes with every path in the diagram still drawn and still useless.
So a mesh topology that has never been tested by failing a link is a diagram, not a design. The links are the easy half.
PitfallsWhere people go wrong
Taking the diagram literally. The picture of computers cabled to each other is teaching fault tolerance, not showing a build. Nobody has nine network cards in a desktop.
Assuming a mesh topology removes single points of failure. It removes them at the layer where the mesh exists. Three tunnels between two sites over one circuit is still one circuit.
Confusing a full mesh with a fast network. Multiple paths is not more speed. It is fault tolerance, and each extra path costs money whether or not it ever carries data.
Meshing everything because meshing sounds resilient. Every added link is another interface, another piece of configuration and another thing to monitor. A partial mesh at the points that matter beats a full mesh nobody maintains.
Buying wireless mesh to fix a coverage problem cabling would fix. Every wireless hop costs throughput, because the radio relays rather than doubling. Where a cable can reach an access point, run the cable.
Reading a mesh diagram without asking which layer. The same network is a mesh at the routing layer and a set of point to point links underneath, and both statements are true.
ComparisonFull mesh, partial mesh, star and hub and spoke, side by side
| Criterion | Full mesh | Partial mesh | Star | Hub and spoke |
|---|---|---|---|---|
| Paths between two nodes | Many | Several, for some pairs | One | One |
| Single point of failure | None | Fewer | The center | The hub |
| Links for n nodes | n(n-1)/2 | Between n and n(n-1)/2 | n | n |
| Interfaces per node | n-1 | Two or three | One | One |
| Adding a node | A link to everything | A link to two neighbors | One link | One link |
| Where you meet it | Core routers, small WANs | Most resilient networks | Every office LAN | Most multi site WANs |
| Usually implemented as | Routes or tunnels | Routes or tunnels | Cabling | Circuits |
The bottom row is the honest summary. The star is cabling, and the meshes are software, which is why comparing them as though they were the same kind of choice is where most explanations go wrong.
FAQFrequently asked questions
What is a mesh topology?
A network in which devices connect directly to multiple other devices, so more than one path for data exists between any two of them. That fault tolerance is the point of the arrangement.
What is the difference between full mesh and partial mesh?
In a full mesh every node connects to every other node. In a partial mesh only selected nodes have multiple connections, which keeps most of the resilience at a small fraction of the cost.
How many links does a full mesh need?
n(n-1)/2 for n nodes. Ten nodes need forty five links, and twenty nodes need one hundred and ninety.
Why is a full mesh so expensive?
Because of the interfaces, not only the links. Each node needs n-1 of them, so a full mesh of ten devices needs nine ports on every device, and ordinary hardware does not have them.
What is the main advantage of a mesh topology?
Fault tolerance, with no single point of failure. If a link or a device fails, data takes another path, which is why mesh networks are used where an outage is more expensive than the extra links.
What are the disadvantages of a mesh topology?
Cost, hardware requirements and configuration effort, all of which grow faster than the number of devices, plus more equipment to monitor and keep consistent.
Where is mesh topology actually used?
Three kinds of network: the routed core of the internet, wide area overlays built from tunnels between sites, and wireless mesh networks where access points relay for each other.
Is mesh WiFi a mesh topology?
Yes, and it is the closest thing to the textbook diagram that exists, because the links really are between the units themselves. The difference is that they are radio links rather than cables.
What is 802.11s?
The IEEE standard for wireless mesh networking. It was issued in 2011, folded into the base 802.11 standard in 2012, and defines a routing protocol so units can find multi hop paths between each other.
Is the internet a mesh topology?
At the routing layer, yes, and it is the largest one there is. The fiber underneath is mostly point to point, which is the general rule that a mesh is a property of paths rather than of cabling.
Does a mesh network make my network faster?
No. It adds fault tolerance. The multiple paths carry data only when the primary path is down or when the routing deliberately spreads load.
What is a partial mesh good for?
Putting redundancy where it is worth paying for and nowhere else, which is what nearly every resilient production network does.
Can I build a mesh topology with ordinary switches?
You can cross connect switches, and spanning tree will then block the redundant links to prevent loops, keeping them ready as spares. The result is a physical mesh with one active path at a time.
Is mesh better than star topology?
They answer different questions. A star topology is how devices attach to a network, a mesh topology is how a network survives a failure, and almost every real network uses both in different layers.
What are mesh topology advantages and disadvantages?
The advantages of a mesh topology are redundancy, no single point of failure and short paths between nodes. The disadvantages are cost and complexity, because a full mesh of n nodes needs n times n minus 1, divided by 2, links. Most real networks therefore use a partial mesh.
Keep readingRelated concepts
Read next · Routing What Is BGP? The protocol that turns the largest mesh anybody has built into a set of routing decisions rather than a set of cables. Open this next12 min- Design · 9 min Hybrid Topology, and Why You Already Have One Why a real network is a mesh in one layer and something else in the others, and why the boundaries between them are the risk.
- Design · 9 min Hub and Spoke Topology, and the Traffic That Goes the Long Way The opposite end of the same dial, with the link arithmetic that decides how much mesh a wide area network can afford.