The 1,024-Client Auditorium
Sales is about to commit to a 1,000-seat auditorium using 12 access points, citing a datasheet figure of "up to 1,024 associated clients per AP." The peak scenario: 1,000 attendees, mixed devices, average 1.5 Mbps per active client during a live-streamed keynote, with 70% of attendees active simultaneously. Each AP is planned on one 5 GHz 40 MHz channel at MCS7 with 2 spatial streams, roughly 300 Mbps PHY and about 180 Mbps TCP goodput; design policy caps real-time traffic at 60% airtime utilization. Thirty percent of client devices are 1×1 phones. Seating is tiered, so client load is never evenly spread across APs.
1. Compute realistic per-AP active-client capacity and compare it to the requirement. Is the 1,024 figure relevant?
Read the worked answer
The 1,024 figure is an association-table limit, not a performance number. It describes how many devices the AP can keep in a table, not how many can be served. It is exactly the wrong number to design to, and the right number to explain in the kickoff meeting.
Airtime arithmetic:
- 300 Mbps PHY → ~180 Mbps TCP goodput at 60% efficiency.
- Mixed device population (30% 1×1 phones) and retries under density reduce that to roughly 140 Mbps of usable throughput.
- Design ceiling of 60% airtime for real-time traffic → ~84 Mbps usable.
- At 1.5 Mbps per active client: 84 / 1.5 = ~56 active clients per AP.
Requirement: 70% of 1,000 = 700 active clients. Twelve APs × 56 = 672 - already short, before any margin, and the calculation assumes an even spread the tiered seating guarantees will not happen. The plan fails on performance while passing on associations, which is the worst combination: it will look fine on paper and collapse in the keynote.
2. What design does work, and how do you present it to sales?
Read the worked answer
To sales, lead with the distinction: the datasheet number is a table size; quoting it converts a technical risk into a contractual one at acceptance testing. Frame the replacement as protecting the deal, not shrinking it.
Design that works:
- 20 MHz channels, not 40. More reuse in a single room and better per-subcarrier SNR at high density; the 40 MHz width is buying spectral efficiency you cannot exploit when 700 clients are contending.
- More APs. 700 / 56 = 12.5 is the floor; with uneven load and 30–40% headroom, plan 18–20 APs on staggered channels, directional antennas aimed down the seating tiers, and reduced power to limit overlap. Note the direct contradiction: more APs at narrower width, not fewer at wider.
- Steer and cap. 5 GHz-only for capable devices, band steering for the rest, low rates disabled.
- Fix the keynote stream. One 1.5 Mbps multicast stream replicated to 1,000 clients is ~1.5 Gbps of airtime and will fail regardless of AP count. Convert multicast to unicast at the edge or use a dedicated distribution method.
If budget caps the AP count, the honest options are to lower the per-client bitrate assumption or to limit concurrent active devices - and to put whichever assumption is chosen in writing, signed by whoever owns the number.
3. The event runs in six weeks and procurement will not fund more than 12 APs. What do you commit to, and what do you refuse to sign?
Read the worked answer
Commit to: a written design basis stating the assumptions - 1.5 Mbps per active client, 70% concurrency, 60% airtime ceiling - and the resulting supported active-client count for 12 APs. With 12 APs the supported figure is roughly 672 active clients at best, and materially less in tiered seating. Recommend 20 MHz anyway, since it costs nothing and improves density behaviour.
Refuse to sign: any acceptance criterion expressed as "supports 1,000 clients" or referencing the 1,024 association figure. Association is not service.
Offer the decision to the business in its own terms: 12 APs will deliver a usable keynote for the subset of attendees active at any one moment - roughly two-thirds of the room - with degraded experience for the remainder. If all 1,000 must be active simultaneously at 1.5 Mbps, the requirement is about 18–20 APs and the shortfall is a scope decision, not an engineering one.
De-risk with measurement: a pilot in the actual room with 200 devices and a synthetic load generator, two weeks before the event, using the same metrics in the acceptance criteria. If the pilot contradicts the model, there is still time to rent additional APs; if the event fails without a pilot, there is no defence.