Six APs, One 2.4 GHz Band, Sixty Users
A 3-floor office with about 90 staff and 60 concurrent devices has six access points, all running 2.4 GHz only - the 5 GHz radios were administratively disabled because 'the laptops are old'. All six APs transmit at maximum power on channels 1, 6 and 11, and channel 6 is used twice: AP-3 and AP-6 sit directly above one another on adjacent floors. 40 MHz channel width is enabled on 2.4 GHz. A survey at busy hour shows 42% frame retries on AP-3, a corner office at -82 dBm with an 18 dB SNR, and users who walk from the meeting room to their desks stay associated with the far AP for up to 15 minutes before roaming. Two 802.11g-only barcode scanners must stay connected. The voice handsets are all dual-band. Every AP is already cabled and no new cabling is available.
1. Rank the three biggest contributors to poor performance and cite the evidence for each from the scenario.
Read the worked answer
Channel reuse and width, not signal strength. 2.4 GHz has only three non-overlapping 20 MHz channels (1, 6, 11). Reusing channel 6 on AP-3 and AP-6, which are vertically adjacent, creates co-channel interference where both APs must take turns on the same medium. Enabling 40 MHz on 2.4 GHz makes it worse: a 40 MHz channel consumes most of the band and overlaps neighbours, which is the direct cause of the 42% retry rate on AP-3.
Maximum transmit power on every AP. High power makes each cell large and asymmetric - the AP can be heard by the client long after the client's low-power radio can be heard well by the AP. That is the sticky-client symptom: the handset clings to a far AP for 15 minutes because the beacon still looks usable. It also increases co-channel interference between floors.
A weak cell edge at the corner office. -82 dBm at 18 dB SNR implies a noise floor around -100 dBm: the signal is weak, not noisy. No channel change fixes that; the coverage hole is a placement/power problem, and low 802.11b rates let clients keep a marginal association instead of roaming.
2. Give the configuration changes that fix this without breaking the two 802.11g scanners.
Read the worked answer
- Enable the 5 GHz radios on all six APs and steer dual-band clients there. 5 GHz has far more channels and is where the handsets and laptops belong.
- Set 2.4 GHz to 20 MHz only, and assign 1, 6, 11 so that vertically adjacent APs never share a channel: for example 1/6/11 on floor 1 and 11/1/6 on floor 2, with AP-6 moved off channel 6.
- On 5 GHz use 20/40 MHz (80 MHz only if channel availability allows) with a clean non-overlapping plan.
- Reduce 2.4 GHz transmit power so cells are roughly equal in size and the 2.4 GHz coverage sits inside the 5 GHz coverage; keep 5 GHz power higher so clients prefer it.
- Enable band steering and set a minimum RSSI (for example -70 dBm) so clients at the far cell edge are pushed to associate elsewhere instead of dragging the data rate down.
- Disable the 1, 2, 5.5 and 11 Mbps rates on the APs serving normal users to shrink the cell and force earlier roaming - but keep them enabled on the single AP that serves the barcode scanners, or give the scanners their own SSID on that AP. Disabling those rates everywhere is what breaks 802.11g-only devices.
3. How would you validate the change, and which numbers would you accept as success?
Read the worked answer
Baseline first, then re-survey the same locations at the same busy hour so the comparison is meaningful.
Acceptance targets:
- Frame retry rate below 10-15% on every AP (from 42%).
- Channel utilisation below 50% on 2.4 GHz and below 60% on 5 GHz at busy hour.
- SNR of at least 25 dB at every claimed coverage location, with -67 dBm or better RSSI for voice-grade areas.
- Roaming completes in under 200 ms with no dropped voice call when walking the meeting-room-to-desk path.
- Both barcode scanners associated and completing a scan transaction on the SSID that still permits 802.11g rates.
- Throughput test (iperf or similar) at three fixed points per floor showing improvement, not just a nicer survey heat map.
Also monitor for a week: client counts per band, retry counters and disassociation reasons. A design that looks good in a one-hour survey but produces band-steering rejections under load has not been validated.