Check out the new USENIX Web site.


Mobility Experiments

In all the experiments described so far, the clients have been stationary. In this section, we consider how the DenseAP system performs with non-stationary clients. Non-stationary clients fall in two categories, nomadic and mobile.

Figure 20: Locations for mobility experiment
\includegraphics[scale=0.4]{figs/nomadicmap.eps}

Nomadic clients move from place to place, but spend significant time being stationary at each place. In corporate WLANs, most non-stationary clients are nomadic clients. A typical example of a nomadic client is an employee who takes her laptop to various meetings. For nomadic clients, the quality of connection they receive when they are "on the move" is less important than the quality of connectivity they receive when they are stationary.

The other type of non-stationary clients are mobile clients. A Wi-Fi VoIP phone user falls in this category. Such clients are rare in current WLANs, but are likely to become more prominent in future [18]. These clients need seamless connectivity as they move. For such clients, metrics such as delay jitter and smoothness of handoff are more important than throughput. Providing good service to mobile clients in a Wi-Fi network is an active topic of research.

Our system can handle both nomadic and mobile clients. We present results for nomadic clients here. The results for mobile clients are omitted due to lack of space, and are available in [22].

Our system periodically determines the location of each client, and triggers reassociation if the client's position has changed substantially. This works quite well for nomadic clients, since the clients are stationary for most of the time. We demonstrate this with the following experiment.

We setup a client at location 1 on the map shown in Figure 20, and carried out ten 2MB TCP downloads from a server on the wired network. We then walked to location 2, waited for two minutes, and carried out the downloads again. We performed a similar experiment at location 3. We repeated the experiment twice: once without mobility support and once with the support enabled.

The median throughput of the downloads, with and without mobility support is shown in Figure 21. We see that with mobility support enabled, the DC correctly reassociates the client at each location, so its performance does not suffer. Without mobility support, the client continues to be associated with the AP near location 1, and its performance suffers at locations 2 and 3.

Figure 21: Performance of nomadic client
\includegraphics[width=0.7\columnwidth]{figs/nomadic.eps}

NSDI-2008