Home › Topics › Wi-Fi Monitoring: What Users Actually Experience › Roaming Behavior

Roaming: Why Phones Cling to Faraway APs (and How to See It)


Every office with more than one access point eventually produces this ticket: "Wi-Fi works fine at my desk, but when I walk to the conference room my call dies." The user swears it is not their phone, because the phone shows full bars the whole way. You check the conference room AP: up, healthy, half a dozen happy clients. Nothing is broken. And yet the complaint is completely real and reproducible. It will keep happening until you understand one under-taught fact about how Wi-Fi actually works.


That fact: when a device moves between access points — a roam — the decision to move is not made by your network. It is made by the device. Your carefully tuned access points are, roaming-wise, just lighthouses: they can shine, they can suggest, but the ship steers itself. And some ships steer like they have somewhere to be, while others cling to the first lighthouse they ever loved until the signal is a rumor.


This guide unpacks that surprise properly. It covers why clients cling and what sticky-client trouble looks like from the helpdesk side. It shows how to actually observe roaming with the logs and counters you already have. It explains the difference between a slow roam and a failed one. And it covers the gentle infrastructure nudges — power, overlap, minimum rates — that make clinging less attractive. Plus the cheapest diagnostic in wireless: a walk down the hallway with a ping running.


In this guide

  • The core surprise: the client decides
  • Why devices cling
  • What stickiness looks like
  • Observing roams with logs and history
  • Slow roams vs failed roams
  • Gentle fixes that encourage better roaming
  • The walk-and-watch test
  • Where to go next

The core surprise: the client decides


Here is the mental model to install. Each of your access points broadcasts its presence many times a second. A client device listens and keeps a private scorecard of the APs it can hear. Entirely according to its own internal logic, it decides when its current AP is no longer good enough and which neighbor to try instead. That logic lives in the device's Wi-Fi chipset and driver. It is different for every manufacturer, often different between models from the same manufacturer. And it is not published, not standardized, and not controllable from your side of the air.


The infrastructure is not entirely mute. Modern APs can hint (advising a client about better neighbors) or assist (making the security part of a roam faster). In some setups they can refuse service to shove a client off. But every one of those is a nudge aimed at a decision-maker that is free to ignore it, and budget devices ignore enthusiastically. The practical consequences are worth internalizing:


  • Two different phones, side by side in the same hallway, can behave completely differently — one hops crisply from AP to AP, the other rides one AP into the ground. Neither is broken; they just disagree about when to leave a party.
  • You cannot configure your way to a specific roaming behavior. You can only shape the radio environment so that good decisions are easy and clinging is unrewarding — which is exactly what the fixes later in this guide do.
  • When a roaming complaint arrives, "which device, exactly?" is a first-class diagnostic question, not deflection. A problem that follows one device model across the whole building is a client personality; a problem that hits every device in one hallway is your coverage design.

Why devices cling


If roaming were free, clients would do it constantly. It is not free, and understanding the costs explains the clinging. To evaluate a neighbor AP properly, a client must scan — briefly retune its radio to other channels to listen. While scanning it cannot send or receive data. Scanning aggressively means interrupting the user's traffic on a hunch. Then the actual move costs a security handshake with the new AP, a window of vulnerability where a mistimed roam can hiccup a call. From the chipset's point of view, the current AP is the devil it knows. The conservative move is to stay put until things get clearly bad. Battery-powered devices are tuned extra-conservative because scanning burns power.


The result is a built-in bias. Most clients use a "leave when it hurts" rule: roam when the current signal drops below some internal threshold. They do not use a "leave when something better exists" rule. A device walking away from its AP experiences a slowly weakening signal that is always just tolerable. So the threshold never quite trips. The device never looks around, so it never notices the excellent AP directly overhead. That is a sticky client: not defective, just loyal past the point of usefulness.


What stickiness looks like


Sticky clients produce a family of symptoms that sound unrelated until you know the cause:


What gets reportedWhat is actually happeningHow to confirm
"Works at my desk, dies on the way to the conference room"The device clings to the desk-area AP through two walls instead of roamingClient history: still associated to the desk AP from the conference room, at a weak signal
"Full bars but everything is slow"Bars show how loudly the client hears the AP; the AP can barely hear the client's weaker radio, and the link runs at crawl-speed data ratesAP-side per-client view: low signal from the client, low data rate, high retries
"Wi-Fi is slow for everyone in this room"One clinging, distant client talks at glacial rates, hogging the shared airtime everyone else must wait behindOne associated client with far weaker signal and lower rates than the rest
"My call blips every time I walk past the stairwell"Roams are happening but slowly — a gap of a second or two on each handoffAssociation timestamps across APs; the walk-and-watch test below

The third row deserves a highlight, because it is how one person's sticky phone becomes everyone's problem. Airtime on a channel is shared, and slow talkers occupy it longest. The road-hog mechanics are laid out in the RF interference guide. And the weak, distant link itself sheds and retries frames constantly. How that turns into the loss and jitter your ping tests see is covered in the Wi-Fi packet loss guide. A clinging client is a small RF disaster on legs.


Observing roams with logs and history


Roaming feels unobservable — the decision happens inside a phone — but the effects are perfectly visible. Every roam leaves fingerprints on your APs: an association event on the new AP and a disassociation on the old one. Line those up by client and timestamp and you have reconstructed the client's journey. Three places to look, depending on your fleet:


  • Client history views. Fleets with a coordinating controller usually offer a per-client timeline: which APs this device visited, when, at what signal. This is the luxury option — the journey pre-assembled. When a "dies in the conference room" ticket arrives, pull the timeline for that device and the story is usually one screen tall.
  • Association logs via syslog. Standalone APs (and controllers too) emit association and disassociation events as log messages. Point every AP's syslog at one collector and you can grep a client's wanderings across the fleet. This costs an afternoon to set up and answers roaming questions forever after.
  • Roam and client counters. Some platforms expose roam counts, roam failures, or per-client signal and data-rate stats over their management interfaces. Where they exist, trend them — a floor whose roam-failure counter climbs every afternoon is telling you when to go walk it.

While you are in the logs, watch for the opposite pathology too: the ping-pong client. It bounces between two APs every few seconds because it sits exactly on the boundary where both look equally mediocre. Each bounce costs a handshake's worth of disruption, so a ping-ponging client feels like constant micro-dropouts. And a desk that produces one is a coverage-design comment, not a device problem. In association logs it is unmissable: the same client address alternating between the same two APs, dozens of times an hour. Sticky clients and ping-pong clients are the two ends of the same dial, which is why the fixes later in this guide aim at the middle.


Here is what a reconstructed roam looks like in centralized logs — one client, two APs, healthy handoff:


10:42:07 ap-fl2-east  client aa:bb:cc:dd:ee:ff associated, signal -58
10:47:31 ap-fl2-east  client aa:bb:cc:dd:ee:ff signal degrading, -74
10:47:44 ap-fl2-west  client aa:bb:cc:dd:ee:ff associated, signal -61
10:47:44 ap-fl2-east  client aa:bb:cc:dd:ee:ff disassociated (roamed)

Thirteen seconds of weakening signal, then a clean jump: the gap between the last event on the old AP and the association on the new one is effectively zero. Compare that healthy pattern with what trouble looks like next.


Slow roams vs failed roams


Not all bad roams are equal, and the logs distinguish them if you know what to read for. A good roam completes in well under a second; on a call you hear nothing. A slow roam takes noticeably longer — a second or several. That time is usually burned in the security handshake with the new AP. This is especially true on networks where each join requires a full authentication round trip to a central server. Users experience a blip: audio drops a syllable, video freezes a beat, then life resumes. In the logs, you see a visible gap between leaving the old AP and being fully associated on the new one.


A failed roam is the ugly one: the handoff falls apart and the client tears down and starts over. That means full association, full authentication, and critically a fresh DHCP conversation. Calls drop entirely; VPNs disconnect; the user often gives up and toggles Wi-Fi. The log fingerprint is unmistakable: a disassociation, a pause, then the complete new-client join sequence. In other words, a DHCP request in the middle of a hallway walk is the reddest of flags, because a mere roam should never need a new address conversation.


Why does this taxonomy matter so much? Because of who notices. File transfers and web browsing shrug off a two-second roam — the transport protocols retry, the page loads a beat late, nobody files a ticket. Real-time traffic has no such patience: voice packets arrive every few dozen milliseconds, and a one-second gap is an audible hole in a sentence. This is why roaming trouble so often surfaces as "calls drop when people walk" while every file-based test you run looks clean. If your organization runs voice or video over Wi-Fi, roaming quality is not a nice-to-have metric. In that case, it is the difference between wireless telephony working and being quietly abandoned by your users.


One genuinely useful infrastructure lever exists here. The standards family includes fast-roaming assists, where APs pre-arrange the security material so a roaming client skips most of the handshake. Where your equipment and clients support it, enabling it turns many slow roams into instant ones. But test with your real device mix before rolling it out broadly. A minority of older clients react badly to the option even being advertised. That caveat is evergreen: every roaming improvement must be tested against the worst device you actually own.


Tip: when a roaming complaint involves calls or video, ask "does it blip or does it drop?" A blip points at slow roams (handshake time); a drop points at failed roams (full rejoin, look for DHCP). One question, asked at the helpdesk, sorts the investigation before it starts.


Gentle fixes that encourage better roaming


You cannot command clients to roam, but you can redesign the incentives. All the effective fixes share one theme: make the far AP less attractive and the near AP more obvious.


  • Turn the loud APs down. The counterintuitive classic. An AP at maximum transmit power is audible three rooms away, so clients cling to it three rooms away. And remember the asymmetry: the AP's big radio reaches the client long after the client's small radio stops reaching back usefully. Reducing AP power shrinks the seductive zone. It pushes the client's "this hurts, look around" threshold to trip nearer the cell edge, where a better AP is actually available. Coverage design is a dial, not a volume war.
  • Design overlap on purpose. Cells should overlap enough that a walking client always hears the next AP at a usable level before the current one degrades into uselessness. That means adjacent APs on different channels, edges meeting at a workable-but-unimpressive signal. Too little overlap creates gaps where there is nowhere better to go; too much overlap re-creates clinging and adds co-channel contention.
  • Raise the floor on data rates. APs advertise the set of speeds they will speak, and by default that set includes ancient, glacial rates. Disabling the slowest rates — a minimum-rate policy — means a distant client simply loses the AP and goes looking for a better one. It does this instead of limping along at a crawl. That minimum-rate policy also shrinks how much airtime any one weak client can burn. Apply gradually and verify with a walk, because an aggressive floor turns marginal areas into hard dead zones. See the dead-zones guide for how to check.
  • Nudge dual-band clients toward the roomier band. Most fleets can encourage a client that supports both bands to prefer the higher, less congested one — typically by answering its probes on that band first. Treat this like every other nudge in this list. It is helpful with cooperative clients, ignored by stubborn ones, and worth testing with your actual device mix. An overeager nudge can slow down joins for devices that hear the higher band poorly.
  • Keep client radios updated. Roaming logic lives in drivers and firmware, and vendors quietly improve it. Where the fleet is managed, keeping wireless drivers current is a legitimate roaming fix, not an IT superstition.

Change one thing at a time, and re-test after each change with the same walk — which brings us to the finest instrument in wireless diagnostics: your legs.


The walk-and-watch test


Every roaming theory meets reality the same way: walk the failing route while watching the device. The method, formally:


  • Step 1. On a laptop (or a phone with a suitable diagnostic tool), start a continuous ping to the default gateway. Open a view showing the currently associated AP identifier (the BSSID) plus signal strength. The identifier is the per-radio hardware address, which tells you exactly which AP you are on.
  • Step 2. Start at the "works fine" location. Note the BSSID and signal.
  • Step 3. Walk the complained-about route at normal speed, watching signal fall and pings tick.
  • Step 4. When the BSSID changes, that is your roam. Note where you were standing and count the missed pings across the transition.
  • Step 5. Repeat the route in both directions, then with a different device type — personality differences between devices are half the diagnosis.

Reply from 10.0.0.1: time=6ms     signal -71, bssid ...:e1 (east)
Reply from 10.0.0.1: time=9ms     signal -76, bssid ...:e1 (east)
Request timed out.
Request timed out.                <-- roam gap: 2 lost pings
Reply from 10.0.0.1: time=7ms     signal -60, bssid ...:w4 (west)

Write down three things per run. Record where the roam happened (mark it on a floor plan) and how many pings it cost. Record the signal level at which the device finally let go. Those three numbers, compared across devices and repeated after each tuning change, are your before-and-after evidence. They turn "it feels better now" into something you can defend in a change review.


Read the results like this: a BSSID change with zero or one lost ping is a healthy roam. Several seconds of timeouts is a slow roam — investigate the handshake path. A roam that only happens when the signal is already dreadful (say, weaker than −80) is stickiness — look at AP power and rate floors. No roam at all along the whole route means one AP is shouting across the entire building. And if the walk shows a spot where no AP is decent, you have found a coverage hole, which is a different disease with its own guide. Twenty minutes of walking converts a vague ticket into one of four specific, fixable diagnoses.


Where to go next


The summary fits on a sticky note. Clients decide. Clinging is rational from where the chipset sits. The fix is shaping the environment — sane power, deliberate overlap, a floor under data rates. The proof is in association logs and a hallway walk. Roaming data earns a permanent slot among your experience metrics (the experience metrics guide shows where it fits). The fleet-side plumbing that collects those logs and counters is covered in the AP health guide. The rest of the series lives at Wi-Fi Monitoring: What Users Actually Experience.


Put this into practice with SysPeep Monitor

Roaming diagnosis leans on history: which AP had the client, when, and what were the radios doing? SysPeep Monitor collects AP and switch metrics over SNMP and receives syslog events. It keeps the dashboards and alert history that let you line up 'call dropped at 10:47' with what the network saw at 10:47. The free trial installs on any Windows machine.

Download Free TrialLearn About SysPeep