Performance improvements for unplanned high density wireless LANs

Chaotic unplanned IEEE 802.11 WLAN deployments are becoming the norm and such residential deployments have many nearby access points (APs) and stations on the same channel, either due to lack of coordination or insufficient available channels. Thus, inter-cell interference in these high-density settings is common but not well-understood. Our evaluations for such interfering deployments reveal that up-to two-thirds of the WLAN system capacity may be lost in a typical large-apartment building with 50 interfering WLANs In this thesis, we first report on our analysis of high-density unplanned WLANs' performance under realistic scenarios. We find that with a typical TCP-dominant workload, cumulative system throughput is characterized by the number of actively interfering APs rather than the number of clients. We verify that due to TCP flow control, the number of backlogged stations in such a network equals twice the number of active APs. Thus, a single AP network proves very robust even with over one hundred clients, while multiple interfering APs lead to a significant increase in collisions that reduces throughput and affects multimedia traffic. Based on our analysis, we suggest a practical contention window adaptation technique, WiPhi, using information on the number of nearby APs rather than clients. We also point out the need for collision-resilient rate adaptation in such a setting. Together these techniques can largely recover the loss in cumulative throughput in a setting with strongly interfering APs. We then propose an alternative ISP-level solution, HeedNet, recovering lost performance by scheduling the IP packets of the bulk traffic at the ISP edge-router towards interfering APs. It requires no changes to the MAC protocol and the APs of the network, making it a viable solution for ISPs. We evaluate HeedNet via simulations and an actual deployment to show that a significant portion of the lost system capacity can be regained (more than 2.2X improvement compared to legacy). HeedNet also increases the fairness, reducing starvation among WLANs. Additionally, we show that HeedNet improves the performance of the non-scheduled (i.e., non-bulk) traffic considerably, such as VoIP, due to the reduced-collision rate environment it creates.Ph.D.Includes bibliographical referencesIncludes vitaby Mesut Ali Ergi