Every ZigBee product page prints a capacity number — 100 devices, 128 devices, sometimes 200. Installers read it as a hard ceiling and budget against it. In practice networks almost never fail because a counter hit its maximum. They fail because routing tables and parent tables inside individual routers quietly fill up, and the symptoms look like random flakiness rather than a capacity problem.
The limit that bites is per-router, not per-network
A mains-powered switch or plug is a router; a battery sensor is an end device that attaches to one parent router and never helps anyone else. Each router keeps a neighbor table and a child table, and those tables are sized in the stack — commonly a handful to a few dozen entries, far below the network-wide figure. When a router's child table is full, a new sensor looking for a parent walks past it and attaches to a weaker, farther router. The network technically supports the device — badly. Everything looks paired, and the link quality tells a different story.
How the failure actually presents
The classic signature: everything works for months, a batch of new devices gets added, and a week later an old sensor in the same room starts dropping offline. Nothing was deleted; the old device did not break. Its parent's tables filled, or a router that powered off and back on lost entries and the sensor never renegotiated. Another signature: after a gateway replacement, devices rejoin the network but sit at terrible link quality, because every end device re-picked a parent from whatever beacon it heard first, not from what was best.
What pushes a table over the edge
Battery-powered devices that join and are never used still occupy a child slot. Motion sensors placed at the edge of range attach to whichever router is barely reachable, consuming that router's table with a link that drops under load. The biggest invisible consumer is churn — a device that re-pairs repeatedly after firmware updates or outages leaves stale entries in some stacks until a restart clears them.
Commissioning rules that prevent it
Place routers before end devices: every switch and plug installed first becomes future capacity. Spread end devices across routers instead of letting them cluster on the nearest one — placing a sensor where two routers both hear it well gives the join process a real choice. Power-cycle routers after large additions so stacks rebalance. Keep a floor plan of parent assignments; it turns "the bedroom sensor is flaky" into "the bedroom sensor chose the hallway plug — move it or add a router there."
What to ask before buying at scale
Ask the platform vendor three things: how large the child and neighbor tables are in the shipped stack, what happens to end devices when a gateway is replaced (full re-pair, network recovery, or automatic), and whether mesh behavior is tunable in firmware. A vendor who answers with numbers has tested the stack; a vendor who repeats the network-wide capacity number has not.
I work on the manufacturing side of this category, and mesh complaints almost never trace back to the radio chip — they trace to stack configuration and placement decisions made on install day. Our smart switch range doubles as router capacity in a mesh, and our documentation as a smart home automation system manufacturer covers gateway and stack behavior in detail. The broader line is at XingLiao and integration questions go through our contact page
Top comments (0)