Discussion

We now discuss some issues related to the DenseAP architceture.

Density Re-visited: The density of DAP deployments affects the performance of DenseAP. This raises some important questions that need to be addressed, (i) Where should the DAPs be placed? (ii) Is there a point at which adding more DAPs to the system can hurt performance? (iii) How do we determine the minimum necessary density for a required level of service in a given environment? Guidelines developed for traditional WLANs offer little help in answering these questions, since they are generally developed with an aim of using as few APs as possible while maximizing the coverage area. Question (i) In our current testbed, we distributed the DAPs roughly uniformly in the given area. However, it may be beneficial to deploy more DAPs near ``hotspots'' such as conference rooms. We are studying this question further. For (ii), thus far, we have demonstrated exploiting density to yield higher gains in capacity. However, with only a finite number of channels and no power control, we expect the benefits from density to diminish beyond a certain point. Mhatre et al. [19] have presented a closed form solution for optimal AP density by varying the CCA threshold, and we are working on validating it on our testbed. To address (iii), we can integrate DAIR [8] with DenseAP to automatically determine RF Holes, i.e. regions with no coverage.

Hidden Terminal: The DenseAP system might exacerbate the hidden terminal problem due to a greater number of parallel transmissions. We have not noticed this effect in our testbed where all DAPs interfere with each other. However, hidden terminals might be a concern in larger testbeds. We are expanding our deployment to investigate this issue in detail. However, our preliminary insight is that the hidden terminal problem might not be severe in the DenseAP scenario because of the capture effect [17]. In a dense deployment of DAPs, the clients are generally located very close to the DAPs they are associate with. Furthermore, the signal in the 5 GHz band fades rapidly in indoor environments thereby reducing the interference from far-away transmitters. Therefore, we expect the capture effect to reduce the impact of hidden terminal problems.

Spatial Reuse of Channels: When assigning a channel to a DAP, our algorithm can take into account the load on all available channels. The load includes background noise, as well as traffic generated by other DAPs. Thus, we achieve spatial re-use whenever possible. Our algorithm, however, is not optimized to maximize spatial re-use.

Co-existence with Other Wi-Fi Networks: Since we can take the load on a channel into account while assigning channels to DAPs, it is easy to see that DenseAP can co-exist with other Wi-Fi networks. For example, if a nearby network is generating heavy traffic on a particular channel, the DenseAP system can detect it, and avoid assigning that channel to DAPs that are likely to be affected by that network.

What is the Ideal Client-AP Assignment?: The ideal client-AP assignment depends on several issues, including traffic, background noise and environmental factors that affect radio signal propagation. Currently, the DenseAP algorithm ignores the impact of hidden terminal issues, and focuses on avoiding problems such as rate anomaly and AP overloading. We make no claims that our algorithm is optimal. In future, we plan to study the optimality of our algorithm using simulations.