Related Work

The conventional AP selection mechanism, based on signal strength measurements, has been shown to lead to poor user experience [1,2,7] and highly unbalanced load distribution among APs [4]. Due to these shortcomings there have been several alternative proposals which typically fall in one of three categories: (i) AP-assisted  [3,7,1], (ii) centralized [2], and (iii) active [8] solutions. In this work we take a step back and look at the fundamental requirements of the AP selection problem. Based on the identified requirements, we propose a technique that does not require the assistance of the AP, does not require previous affiliation of the client with an AP, and is initiated by the client without the need for central coordination. Such properties allow for the continuous evaluation of the ``quality'' of all APs within range that could also facilitate better roaming decisions.

Our work targets the estimation of the potential bandwidth and not the available bandwidth as in [5], which is defined as the maximum rate at which a host can send its data without lowering the sending rates of other already affiliated hosts. In this work, we are not interested in the bandwidth available to a client before affiliation, but the MAC-layer bandwidth the client will receive after it affiliates with the AP. In addition, we do not aim to estimate the layer-3 throughput that a client would receive once affiliated with an AP, since such an estimation would require knowledge of the client's workload and its path through the wired network. The metric of potential bandwidth can characterize the wireless part of the client's connections. In future work, we intend to look into passive measurement techniques that could allow us to extend our estimates to account for the wired part of the network, say by passively observing the performance currently experienced by other users in the same wireless network.

The closest recent work to ours is [6], where the authors propose a methodology for passive bandwidth estimation between two communicating wireless stations. However, their method does not provide an estimate of the potential bandwidth that an end-host is likely to receive on a wireless link with another host (when one of the hosts is not part of the network yet).