Roughly half on a single-radio mesh, considerably less on dual or tri-radio designs with a dedicated backhaul radio.

A single-radio node receives a frame and retransmits it on the same radio and channel. The airtime is used twice for one frame's worth of delivery, so throughput halves. Two hops leaves about a quarter, three about an eighth - the fall is steep and it compounds.

A dual-radio node uses one radio to serve clients and another for the backhaul, so the two do not compete. Loss per hop is then much smaller, though not zero, since the backhaul radios still share the medium with each other.

Tri-radio designs add a second backhaul radio and reduce it further.

The design implication is the same in every case: keep the hop count low. One hop is usually fine, two is workable with dual-radio hardware, and three or more should prompt a hard look at whether one more cable could be run.

When cabling is genuinely impractical, temporary, or disproportionately expensive - not when it is merely inconvenient.

Good cases are clear: listed and heritage buildings where routes cannot be cut; outdoor areas with no ducting; construction sites and events where the network exists for weeks; extending coverage across a yard where the trench cost exceeds the value; and temporary expansion into a space that will be given up.

Mesh is also useful as resilience - a node with a wired uplink that can fall back to a mesh path if the cable fails.

Where it is the wrong answer is a permanent installation in a building where cables could be run. The performance cost is real and permanent, and the saving is a one-off.

A useful test: if the node will still be there in five years and a cable could reach it, run the cable. Mesh is at its best where the alternative is nothing at all.

By a routing metric that weighs link quality and hop count, recalculated continuously as conditions change.

Each node evaluates the paths available to it toward a wired node, scoring them on signal quality, achievable data rate, current load and the number of hops. It selects the best and switches if something better appears or the current path degrades.

That self-healing is the main operational benefit: if a node fails or a path is blocked, neighbours reroute automatically without intervention.

The algorithms are vendor-specific, which is why mesh nodes from different manufacturers do not form a mesh together even though they are all wireless.

The practical implications are to provide more than one viable path where resilience matters, to avoid designs where one node is the only route for many others, and to check what path each node has actually chosen after installation - the management interface shows it, and it is occasionally not what was intended.

No. Throughput falls with each hop and latency accumulates, so practical meshes stay shallow and add wired nodes instead.

Each additional hop reduces the bandwidth available to everything beyond it and adds delay, which affects voice, video and interactive applications noticeably.

Control overhead also grows: nodes exchange routing information, and a large mesh spends an increasing share of its airtime managing itself.

The practical answer to extending coverage is to add wired uplinks rather than depth. A mesh of twenty nodes, each one hop from one of five wired nodes, performs far better than twenty nodes chained from one.

That is the design pattern to aim for: as many wired injection points as can be achieved, with mesh filling the gaps between them at one or two hops.

Where a long distance must be covered with no cabling at all, a point-to-point bridge to a distant wired node, with local mesh beyond it, is usually better than a chain of mesh hops.

A band separate from the one serving clients, and on outdoor meshes usually the higher bands where there is more spectrum.

On a dual-radio node, the client radio and the backhaul radio should be on different bands so they do not contend. A common arrangement serves clients on 2.4 GHz and backhauls on 5 GHz, or serves on 5 GHz and backhauls on 6 GHz where available.

Higher bands offer more channels, so backhaul links between adjacent node pairs can use different channels and avoid interfering with each other - which matters as soon as there is more than one hop in an area.

Outdoor meshes also benefit from directional antennas on the backhaul, concentrating energy along the path between nodes rather than radiating in all directions. That improves the link and reduces interference with everything else.

Check local regulations for outdoor use: some channels are restricted to indoor use or subject to radar avoidance, which can cause a backhaul to change channel unexpectedly.

It is one of the main uses - large yards, car parks, campuses and construction sites - provided the hop count is controlled and the hardware is properly rated.

Outdoors there is often no cabling infrastructure at all, so mesh may be the only practical method of covering an area quickly.

The conditions for it working well are: outdoor rated hardware with proper ingress protection and surge protection; directional backhaul antennas between nodes; clear line of sight between mesh neighbours, since outdoor obstructions such as vehicles and stacked materials move; and as many wired injection points as the site allows.

Power is the other constraint. A node needs power even if it needs no data cable, and PoE from a distant switch may exceed the cable limit. Solar and battery nodes exist for exactly this and change the power budget conversation entirely.

Expect to revisit the design as the site changes - on a construction site in particular, what had line of sight in month one may not in month six.

Through the controller's mesh view: each node's parent, hop count, backhaul signal quality and throughput - and those four figures explain most problems.

A node whose hop count has increased has lost its preferred parent and is routing the long way round. That is usually the first sign of a failing node or a new obstruction.

A node whose backhaul signal has fallen has an alignment, obstruction or interference problem on that specific link.

A node serving clients well but delivering poor throughput is usually too many hops from the wire, which is a design issue rather than a fault.

Record the intended topology at installation - which node should parent to which - so deviation is visible. Meshes reconfigure themselves silently, and a mesh that has quietly rearranged itself into a chain is a recurring source of a gradual decline that nobody can attribute to any particular event.