Overall DenseAP Performance

Figure 10: DenseAP performance: 802.11a

Figure 11: DenseAP performance: 802.11g

We now repeat the experiment described in the previous section, but using the DenseAP system instead of corporate WLAN. All features such as channel assignment, association policies and load balancing were enabled. We repeated the experiment twice in 820.11a mode, once using 8 channels (channels 36-64) and once using 4 channels (channels 40, 48, 56 and 64) We also ran the experiment with the DenseAP system in 802.11g mode using 3 orthogonal channels (channels 1, 6 and 11).

Figure 10 illustrates the performance of DenseAP in the 802.11a band. Figure 11 illustrates the performance of DenseAP in the 802.11g band. The graph does not have a baseline, since we do not wish to compare performance of corporate WLAN and DenseAP in 802.11g mode, as explained earlier. Let us focus on the 802.11a results.

We see significant performance gains over the corporate network. For example, with 8 simultaneously active clients, the median download throughput on the corporate network was 1.3Mbps. On the other hand, the median download throughput with DenseAP when using 8 channels, was 11.25Mbps. This represents an improvement in capacity by a factor of 868% over the corporate WLAN. Similarly, for 12 clients in the system, the median download throughput for corporate WLAN is 750 Kbps and for DenseAP it is 9.4 Mbps, which is an improvement of over 1250%.

The the comparison with the corporate WLAN may seem unfair, because we are comparing the 8-channel, 24-AP DenseAP system against a single-channel, single-AP baseline. However, the only purpose of these results is to show the full benefit of the DenseAP approach in our testbed. The next step is to separate out the impact of various factors that contribute to these results. As described earlier, the gain in throughput comes from four factors. These are: (i) use of orthogonal channels (ii) dense deployment of APs (iii) use of intelligent associations and (iv) load balancing.

We note that though enabled, the load balancing algorithm played no role in these results. The main reason is that the clients are scattered uniformly across the floor. Thus, in most cases, each client associated with its own DAP. Further, since all clients started at the same time and they all saturated their respective channels, there was no opportunity for our load balancing algorithm to move a client from one DAP to another since all channels were equally loaded. We consider the impact and efficacy of the load balancing algorithm later in Section 6.3.

It is easy to see that more orthogonal channels are better, since the median throughput is higher with 8 channels than with 4 channels. But the important question is whether the DenseAP system derives all its benefit from using more orthogonal channels? That is, can we isolate the impact of the dense deployment of DAPs and our centralized association policy?

To isolate the impact of DAP density, we need to ensure that the number of channels and the association policy play no role in the performance. The way to do this is to evaluate the performance of the DenseAP system with all DAPs operating on the same channel. This experiment is described next.