The Available Capacity Metric

When we select a DAP for a client to associate with, we want to pick a DAP where a client can expect to get good throughput. Unfortunately, it is not easy to determine what throughput a client can expect to get when associated with a DAP. The value depends on several factors, such as quality of the channel between the DAP and the client, presence of other traffic/interference, autorate algorithms in use and the CCA thresholds used by the client and the AP. Rather than estimating each of these factors, we focus on the two that affect the expected throughput the most, namely: the transmission rate the client and the DAP can use to communicate with each other; and how busy the wireless medium is in the vicinity of the client and the AP.

The transmission rate is a function of, among other things, how well the signal propagates between the client and the DAP. For example, if a DAP and a client are far away from each other, they generally won't be able to communicate at high transmission rates, since the wireless signal degrades with distance. Alternatively, when a DAP and a client are close to each other, they will be able to communicate at higher transmission rates. However, if the wireless channel is busy with other traffic, the expected throughput will be lower, since the client and the DAP will have fewer opportunities to transmit packets.

The combined impact of the transmission rate and the busy medium is approximated by the Available Capacity ($AC$) metric as follows. Given a channel ($C$), a DAP ($D$) and a client ($M$), $AC^C_{DM}$ is the product of free air time on $C$ in the vicinity of $D$ and $M$, and the expected transmission rate between the $D$ and $M$. The free air time is simply the percentage of time when the wireless medium is not in use. Our notion of load at a DAP on a particular channel is defined by (1 - free air time).

The intuition behind this metric is that the DAP with the highest available capacity will allow the client to send the most data, while simultaneously reducing the impact on other clients in the network. For example, if a client and DAP can communicate at a high transmission rate, then each frame will consume less air time, and the client will be able to send more data. Furthermore, other clients on the same channel in the vicinity will have more opportunity to transmit. On the other hand, if the channel has low utilization then it is acceptable for the client and the DAP to communicate at a low transmission rate because it will have little impact on other clients.

This metric is similar to the one proposed in [10], where clients associate with the AP which is the least loaded and offers the best data rate. We compare and contrast our work with [10] in more detail in Section 8.

We now describe how we estimate the free air time and the expected data rate for a given DAP/client pair. We stress that we do not expect these calculations to be precise. Our intention is to provide a reasonable ordering of DAPs, in order to pick a good AP for the client to associate with. The load balancing process described later in the section can reassign the client to a different DAP, should the conditions change and the initial choice is no longer appropriate.