Searcharxiv⌕ Search

arXiv subjects

Paul Patras

Publications and source records attributed to Paul Patras.

20 records · Page 2Linked to original sources

Rigorous and Practical Proportional-fair Allocation for Multi-rate Wi-Fi

Recent experimental studies confirm the prevalence of the widely known performance anomaly problem in current Wi-Fi networks, and report on the severe network utility degradation caused by this phenomenon. Although a large body of work addressed this issue, we attribute the refusal of prior solutions to their poor implementation feasibility with off-the-shelf hardware and their imprecise modelling of the 802.11 protocol. Their applicability is further challenged today by very high throughput enhancements (802.11n/ac) whereby link speeds can vary by two orders of magnitude. Unlike earlier approaches, in this paper we introduce the first rigorous analytical model of 802.11 stations' throughput and airtime in multi-rate settings, without sacrificing accuracy for tractability. We use the proportional-fair allocation criterion to formulate network utility maximisation as a convex optimisation problem for which we give a closed-form solution. We present a fully functional light-weight implementation of our scheme on commodity access points and evaluate this extensively via experiments in a real deployment, over a broad range of network conditions. Results demonstrate that our proposal achieves up to 100\% utility gains, can double video streaming goodput and reduces TCP download times by 8x.

cs.NI↗

Control Theoretic Optimization of 802.11 WLANs: Implementation and Experimental Evaluation

In 802.11 WLANs, adapting the contention parameters to network conditions results in substantial performance improvements. Even though the ability to change these parameters has been available in standard devices for years, so far no adaptive mechanism using this functionality has been validated in a realistic deployment. In this paper we report our experiences with implementing and evaluating two adaptive algorithms based on control theory, one centralized and one distributed, in a large-scale testbed consisting of 18 commercial off-the-shelf devices. We conduct extensive measurements, considering different network conditions in terms of number of active nodes, link qualities and traffic generated. We show that both algorithms significantly outperform the standard configuration in terms of total throughput. We also identify the limitations inherent in distributed schemes, and demonstrate that the centralized approach substantially improves performance under a large variety of scenarios, which confirms its suitability for real deployments.

cs.NI↗