It is difficult to accurately predict the transmission rate a client will achieve when communicating with a DAP (or vice-versa). The rate primarily depends on how well the DAP receives the client's signal. However, the rate also depends on a variety of other factors such as the autorate algorithm implemented by the client, power levels used by the client, and channel conditions near the client. Of these factors, we can only estimate how well the DAP receives a client's signal.
When attempting to associate, clients send out probe request messages which are overheard by nearby DAPs who then inform the central controller. We estimate the quality of the connection between the client and the various candidate DAPs using the signal strength (RSSI) of the received probe request frames at the various DAPs. We convert these observed signal strengths into estimates of expected transmission rate by using a mapping table. The mapping table bucketizes RSSI values into fixed-size buckets, and assigns an expected rate to each bucket. We assume that the same transmission rate will be used by both the client and the AP. We call this the rate-map approach. The mapping table is initially generated by manual profiling using a few clients at various locations. It can then be refined as actual data from more clients is gathered during live operation.
At first glance, it may appear that extrapolating the signal strength observed in the uplink direction to an expected transmission rate in both directions could result in inaccurate estimations and/or poor performance, especially considering the other factors that are ignored. Yet, in our system, we find that it provides reasonable results for the following reasons. First, given the density of access points, a client generally associates with a nearby DAP. For such short distances, we find that signal strength measured in one direction is a good approximation of signal strength seen in the other direction. Second, because the client and the DAP are usually close to each other, we generally see good signal strength in both directions. Most commercial Wi-Fi cards behave similarly in such conditions. Finally, note that we do not need the exact transmission rates used by either the client or the DAP. The conversion table is merely a way to ranking the relative importance of the observed signal strength. In Section 6, we will present results that demonstrate the usefulness of the rate-map approach.
We now describe how the AP selection algorithm uses the available capacity metric.
NSDI-2008
