Distribution Center: Wide Channels, High Retries
A 18,000 m² automated distribution center runs 24 ceiling-mounted 5 GHz access points on 40 MHz channels. Coverage surveys passed, but since go-live the 220 handheld scanners show a 34% average retry rate and 68% channel utilization at peak (10:00–14:00), with 30–45 disconnects per minute clustered at aisle intersections. The scanners are 1×1 and 20 MHz-capable only. Operations wants to raise AP transmit power "so the signal is stronger"; the integrator proposes adding 12 more APs inside the same 40 MHz plan. There are no 2.4 GHz devices on site and no neighbouring 5 GHz networks. A four-hour Sunday maintenance window is available. Peak-hour downtime is estimated at about $9,000 per hour.
1. Of the three proposals - raise transmit power, add 12 APs on the same 40 MHz plan, or re-plan channels - which is defensible? Justify it from the measurements and specify what you actually change in the four-hour window.
Read the worked answer
The measurements describe contention, not coverage. A 34% retry rate means roughly one frame in three needs retransmission, and 68% utilization means the medium is busy two-thirds of the peak hour. Coverage was never the complaint.
- Raise power - no. Expanding each cell's footprint increases co-channel overlap and makes scanners stick to distant APs. It addresses a problem that has not been observed and worsens the one that has.
- Add 12 APs at 40 MHz - no. That puts 36 radios into roughly 12 usable 40 MHz channels in most regulatory domains, raising co-channel interference on a medium already 68% busy. It also buys nothing for the clients: a 1×1, 20 MHz-only scanner cannot use the extra 20 MHz of width.
- Re-plan channels - yes. Moving to 20 MHz roughly doubles the available non-overlapping channels (~24–25 versus ~12), which makes a near 1:1 reuse of 24 APs possible instead of an average of two APs per channel.
In the window: (1) re-plan to 20 MHz and assign channels so no two APs sharing a channel can see each other, staggering across aisles; (2) reduce transmit power so each cell covers its own aisle span; (3) enable 802.11k/v/r and a minimum-RSSI disassociation threshold to force roaming at the intersections; (4) disable low and legacy rates; (5) capture a 30-minute post-change baseline. Save configs first - rollback is a restore, about 15 minutes. Because full disruption costs roughly $36,000 across four hours, change one zone at a time and verify before moving on.
2. What is your success criterion at the end of the window, and which metric will you refuse to accept as proof that the fix worked?
Read the worked answer
Accept: retry rate below 10% (from 34%) and disconnects below 5 per minute (from 30–45), measured in the same time-of-day peak on the same client population; utilization below 50% as a secondary signal.
Refuse: throughput. A single scanner tested alone will show attractive throughput on a badly contended design, because throughput is a lagging aggregate that hides airtime unfairness. The scanners' complaint is disconnects and retries, so those are the metrics that must move. Also refuse raw signal strength - a high RSSI is exactly what the power increase would have bought, and it is what caused sticky clients.
The honest problem with the window: Sunday has no production load, so you cannot validate a contention fix at 10:00 Monday conditions on Sunday. Plan a controlled load test with ~40 scanners in one zone during the window, and schedule a named owner to watch Monday 10:00 with a pre-authorised 30-minute rollback trigger.
3. Operations argues that at $9,000 per hour of downtime you should skip the Sunday change entirely and add the 12 APs during production week instead. Respond.
Read the worked answer
The instinct to avoid a disruptive window is right; the chosen alternative is worse.
- Adding APs in production is the highest-risk option available. You would be changing RF mid-shift with no baseline, no rollback plan and no way to attribute a regression. In a contention-bound design, more radios at 40 MHz make the measured symptom worse, so you would pay the risk and not get the benefit.
- The window is config-only. No hardware, no truck roll, saved configs, restore in about 15 minutes. That is the cheapest failure mode you will get.
- The real cost is not the full four hours. Phasing zone-by-zone limits exposure to one zone at a time rather than the whole facility, so the $36,000 figure is a worst case for a change you would never do all at once.
Concede the valid part of the objection: the Sunday window cannot validate under real load. Pair it with one weekday peak-hour observation with a rollback trigger, so you are not declaring success on a quiet afternoon.