* joelja@bogus.com (joel jaeggli) [Sun 16 Sep 2012, 18:42 CEST]:
We tend to engineer for a maximum of around 50 associations per radio (not AP). beyond that performance really starts to suck which can be measured along a multitude of dimensions. The most visible one to the client(s) being latency due to loss and subsequent retransmission.
Reduction in coverage is done on a couple of dimensions. that ap with the 3-5dBi gain dipoles probably shouldn't be 100mW. but the noise floor is in a different place when the room is full of clients so it can't be to low either. Dropping the low speed rates backward compatibility with 802.11b and setting the multicast rate to something higher will force clients in marginal coverage situations to roam more quickly, hog the air less and allow for higher throughput.
This is all good advice that you should implement. The difficulty with high user density deployments is getting stations to associate to the nearest access point on the optimal band. When presented with the same SSID for 2.4 and 5 GHz, clients usually prefer te 2.4 GHz one because its S:N ratio usually seems better (inherent to the lower frequency). However, in practice this isn't always the case as there usually are many more clients on 2.4. Various vendors of lighweight access points use tricks to get clients to associate on the 5 GHz band: e.g. Cisco, I think, will reject an initial association request at 2.4 GHz in the hope that the client will retry at 5 GHz before retrying at 2.4, which will both be accepted. -- Niels.