Skip to content

Clients

Every device seen on the network, where it is connected, and what it has been doing.

Where the data comes from

Clients are assembled on the fast poll from several sources on each managed device: DHCP leases, the ARP table, the bridge host table, and wireless registrations. A single client usually appears in more than one of these, and they are merged by MAC address.

That merge is why a client shows a switch port and an access point: the MAC is known to both.

When several devices know a client, it's shown where it's connected: the access point it's registered to, or else the switch that sees it on an edge port. Every switch between the client and the router learns its MAC, most of them on an uplink, so a sighting on a port facing another managed device (an uplink, bond or peer link) comes last (#363).

A sighting from the last day counts for where a client is, even if it has gone quiet: overnight a switch forgets an idle PC while the router's ARP table still lists it, and the PC stays on its switch port rather than moving to the router. Whether a client is online, and when it was last seen, come from all of its sightings: online if any device sees it now.

An ARP entry only counts while the router is actually in touch with the client: reachable, delay, probe, or a static entry. RouterOS keeps stale and failed entries, MAC and all, until it gets round to clearing them, which on a quiet VLAN can take days, so counting them kept phones online all weekend after they'd left (#363). A failed entry never counts. A stale one keeps a client that's already online there for 15 minutes after the router last saw it properly, since a printer or a phone on standby that's still there spends most of its time stale, but it doesn't bring a client online, and it doesn't move Last seen.

On the client's page, Connection Path shows the device the client is on and the whole way from there to the top of the site, a step at a time, with the port each step leaves by and the port it arrives at. Names wrap in full and link to the device.

Each step is spanning tree's answer where it has one, the same one the topology map uses: the switch at the other end of the device's root port, managed or not. An unmanaged switch is shown by the name discovery heard, marked not managed, and the path ends there. A device that is the root bridge is the top of its tree, so from there the path only goes up a tier, to a router above a switch or a switch above an access point, never back down into the tree. An access point whose bridge hears no BPDUs calls itself root, and still goes up to its switch.

Where spanning tree has no answer, the step comes from neighbour discovery: a link on a bridge port (or a member of a bonded one) rather than an out-of-band management port, and among those a neighbour heard over LLDP. Bridges pass MNDP and CDP on but not LLDP, so an access point behind a switch also hears the router through it, and only the LLDP neighbour, the switch, is at the other end of the cable. When links still tie, a router comes before a switch, and a switch before anything else, and a neighbour whose own way up is this device comes last (#365). A neighbour the manager doesn't manage is a step only when there's no managed one, and only if it was heard over LLDP. The top of a site has no way up: when the best neighbour goes up through this device and is a tier below it, the path stops. Hover an arrow to see which way chose it.

A path can cross into another site. Discovery only matches a neighbour to a device in the same site, so a switch cabled to one managed in another site looks unmanaged from here. The path still shows it as that managed device when the neighbour proves it is: its MAC is one of that device's own, no other device in the fleet has it, and it calls itself by that device's name. The path then goes on in the other site. If you can't see that site, the step shows the name discovery heard, marked in a site you can't see, and the path stops there.

A client seen only on uplinks, behind equipment the manager doesn't manage, is placed on the device deepest in the tree that sees it on a port leading down, rather than on one that sees it on its way up (its root port, or a link to a device a tier above).

A managed device's own address is heard by every switch around it. Its page says so, links to the device, and starts the path there; it used to start at a neighbour that heard it, and lead back to the device itself.

This can be switched off

Client tracking is the heaviest collector on most fleets. Settings → Polling → Collectors can disable it, at the cost of this page and everything fed by it. See Polling and scaling.

The list

Network devices aren't clients, so the list leaves them out: the MAC addresses of managed devices (any interface or radio), VRRP's virtual MACs, and switches, routers and access points the manager doesn't manage but that announce themselves over LLDP, CDP or MNDP as a bridge, router or WLAN access point (#363). An IP phone that also says it's a bridge, for its PC port, stays a client. The filter bar says how many are hidden, with show to bring them back.

Filter by status (online, or all including inactive clients), type (wired or wireless), the device a client is connected to, VLAN, or signal range, in the same filter bar as the device list. Search matches hostname, MAC, IP address and vendor. Refresh, the automatic refresh interval and Purge stale sit beside the page title. Refresh asks the devices in view for a fresh poll and reloads the list as their answers arrive over the next few seconds; the Wireless and Routing pages' Refresh works the same way.

Every column sorts. With Wireless selected, SSID, Signal and Rate replace the Type column and sort too; Signal sorts strongest first on the first click. Rate is the client's current link rate, seen from the access point: TX is towards the client, RX from it. It's also on the client's page, and comes from the newer Wi-Fi package, the legacy wireless driver and legacy CAPsMAN alike.

The list has a minimum width, so in a narrow window it scrolls sideways rather than cutting off IP addresses (#336). Hover a shortened name, vendor, SSID, interface or device to see all of it.

Columns at the end of the filter bar shows or hides columns, apart from the host (#336). Drag a column header's right edge to make it wider or narrower, and double-click that edge to put it back. Reset columns restores the defaults. The choice is remembered in your browser and applies in every view: a hidden column stays hidden whether you're looking at wired, wireless or all clients.

Active means the client was present at the most recent poll of the device it is attached to. Clients that disappear are kept and marked inactive rather than deleted, so history survives a laptop closing its lid; retention_clients_days (default 7) decides how long an inactive client is kept.

Identification

Each client is classified into a category — network gear, server, computer, phone, TV, camera, printer, game console, voice assistant, smart home, IoT — derived from its MAC vendor prefix and hostname.

This is a guess, and it is sometimes wrong. Both the category and the display name can be overridden per client, and an override persists across polls rather than being re-derived. Use custom name for the label you actually recognise the device by; the hostname it advertises is kept alongside.

Client detail

Selecting a client opens its own page:

Section Shows
Connectivity Where it is attached now, and on which port or radio
Presence When it has been seen, as a timeline
Traffic Per-client throughput history
Signal For wireless clients, RSSI over time
Roaming Movement between access points, with the signal at each

Roaming and signal history are the two that answer questions nothing else can: a client that keeps dropping is usually either roaming between two APs that both hear it weakly, or sitting at the edge of one.

Signal quality bands are −70 / −55 / −45 dBm. A phone typically drops off a network around −75 dBm, so anything in the Poor band is at or past the edge of usable. Dedicated point-to-point radios are an exception and will read Poor while working perfectly.

Per-port clients

The Ports tab of a switch shows who is physically on each port. Uplink and trunk ports are detected automatically — by an LLDP neighbour, by MACs spanning several VLANs, or by a high MAC count — and show an explanation rather than listing every MAC reachable through them. The full table is one click away if you want it.

Actions

  • Address lists (#145), on a client's page next to its IP address, or the list icon on its row: the router's firewall address lists, ticked where the client's address is already a member. Tick or untick, or type a new list name, then Apply. An address can be in several lists at once.
    • Entries a firewall rule added (dynamic) are shown but can't be changed here.
    • Lists a firewall rule matches on are marked used by rules. Every change runs through Change Guard, which asks before anything it predicts would lock the manager out.
    • If the address is a dynamic DHCP lease, a warning says so: the lists follow the address, not the device. Make the lease static to keep them together.
  • Disconnect (#338), on a wireless client's page or the Wi-Fi-off icon on its row, removes it from its access point's registration table, as Remove does in WinBox, so it reassociates, normally within seconds. It works with the Wi-Fi package, the legacy wireless driver and legacy CAPsMAN, and is for operators and admins in the client's site. If the client is one of your managed devices (a wireless backhaul in station mode, say), it asks again first, since that drops the device and everything reached through it until it reconnects. Each disconnect is in the audit log.
  • Wake-on-LAN sends a magic packet from the MikroTik nearest the client, not from the manager, so it reaches devices the manager cannot route to.
  • Notes are free text kept against the MAC.