Benefits of Association Policy

Figure 13: Benefits of the association policy

Figure 14: Gains over the client-driven approach

In this section we demonstrate the benefits of the DenseAP association policy over the client-driven method of association used in conventional WLANs.

We carried out the following experiment in the 802.11a band, using 8 channels. We disabled the DC. We first assigned channels to all DAPs using the channel assignment algorithm described in [15]. We then disabled the ACLs, and allowed clients to associate with the DAP of their choice. In other words, the association decisions were left to the clients (as it is in today's WLANs). This setup represents a dense deployment with a conventional WLAN approach. We then carried out the experiment described in the previous section.

Figure 15: A conference room scenario

We note a few points about this particular scenario. There was no pre-existing traffic on any of the channels. Also, the clients were generally evenly distributed across the testbed, and so were the DAPs. Each client then picked the DAP to associate with based on the local client driver implementation policies.

The results of the experiment are compared with the result of running a full fledged DenseAP system with the same deployment and 8 channels. These results, along with the baseline, are shown in Figure 13. The results show that while simply deploying more APs and doing intelligent channel assignment in a conventional WLAN will be beneficial, the benefits will be higher if associations are controlled in a centralized manner.

In other words, the fact that the the line labeled ``DenseAP'' is above the line labeled ``Client-driven'' is what demonstrates the benefits of the DenseAP approach. The extra gain is due to the intelligent, centralized association control used in the DenseAP system. The magnitude of the extra gain is illustrated in Figure 14. In fact, as we shall see later, the centralized controller can provide roughly the same gains with fewer APs.

To drive home the point about the benefits of association policy, we consider which DAPs the clients associated with when left to decide by themselves. For example, in the case of 12 active clients, the clients used only 6 channels and 10 APs. On the other hand, by using the association policy, the DenseAP system used all 8 channels, and 11 APs.

One may argue that in the above experiment, the ``Client driven'' approach performed worse than the DenseAP approach simply because the specific static channel assignment we used for the ``Client driven'' approach was a bad one. However, we note that any static channel assignment algorithm that does not take into account the actual location of clients in the system, is always likely to underperform a dynamic, on-the-fly channel allocation algorithm. We demonstrate this with a simple experiment.

We set up three clients in a small conference room, as shown in Figure 15. There were no other clients in the system. We disabled DC and instead let the three clients pick the DAP to associate with. Unsurprisingly, they all associated with the AP located in the conference room. Note that no static channel assignment algorithm can remedy this situation: the clients must associate with different DAPs for channel assignment to have any impact. We repeated the same experiment with the DC enabled and the association policy ensured that the three clients associated with three separate DAPs.

Note however that the association policy alone is not effective. It delivers an improvement in capacity in conjunction with a higher density of DAPs. To demonstrate this, we consider the performance of the system with fewer DAPs.