Identifying High Throughput Paths in 802.11 Mesh Networks: a Model-based Approach Theodoros Salonidis Thomson Paris Research Lab Michele Garetto Universit` a di Torino, Italy Amit Saha Tropos Networks, CA Edward Knightly Rice University, Houston, TX Abstract— We address the problem of identifying high through- put paths in 802.11 wireless mesh networks. We introduce an analytical model that accurately captures the 802.11 MAC protocol operation and predicts both throughput and delay of multi-hop flows under changing traffic load or routing decisions. The main idea is to characterize each link by the packet loss probability and by the fraction of busy time sensed by the link transmitter, and to capture both intra-flow and inter-flow interference. Our model reveals that the busy time fraction experienced by a node, a locally measurable quantity, is essential in finding maximum throughput paths. Furthermore, metrics that do not take this quantity into account can yield low throughput by routing over congested paths or by filtering-out non-congested paths. Based on our analytical model, we propose a novel routing metric that can be used to discover high throughput path in a congested network. Using city-wide mesh network topologies we demonstrate that our model-based metric can achieve significant performance gains with respect to existing metrics. I. I NTRODUCTION Mesh networks offer inexpensive wireless coverage over large areas via use of wireless multi-hopping to wireline gateway nodes. Recently, cities are expanding use of mesh networks from public service and public safety to also in- clude large-scale public broadband wireless access, potentially serving millions of users. 1 Such deployments will carry high amounts of traffic that will stress the 802.11 mesh backbone, thus causing unfairness and starvation, well known problems of the 802.11 CSMA protocol. Given this limitation, modeling and understanding 802.11 in conjunction with congestion con- trol, traffic engineering and routing schemes is of paramount importance. In this paper, we address the problem of identifying high throughput paths in an 802.11 mesh network by introducing a model that accurately predicts throughput and delay of multi- hop flows under fixed or changing traffic conditions. Existing models for 802.11 mesh networks focus on predicting through- put of a set of single-hop, single-receiver flows [6], [11], [13], [14]. A recent paper [10] proposes an analytical approach to estimate the end-to-end throughput over a single path, which is limited to the case of nodes having a single receiver and requires a rather complex, centralized computation that cannot be translated into a routing protocol. In contrast to [10], our model applies to arbitrary traffic matrices and yields a routing metric that can be easily incorporated to an efficient routing protocol to discover the optimal path. In [15] the authors propose an admission control schemes for flows in a single- channel, multi-hop network based on knowledge of both local 1 See Houston’s RFP for example: www.houstontx.gov/it/wirelessrfp.html. resources at a node and the effect of admitting the new flow on neighboring nodes. In contrast to [15], our approach is based on a more precise mathematical model of the behavior of 802.11, and can efficiently discover the path providing the largest bandwidth to the new flow. Our model expresses the throughput and delay of each link of a multi-hop flow as a function of (i) average input rate, (ii) the fraction of busy time carrier-sensed by the link transmitter and (iii) the packet loss probability experienced on the link – all locally measurable with zero or minimal communication overhead. To model changing traffic conditions, we introduce a two-step technique to estimate available path bandwidth, defined as the maximum additional rate a flow can push before saturating its path. The first step computes the capacity of each link in the path using the busy time fraction and packet loss probability as a summary of the interference caused by other links outside the path. In the second step, the link capacities are coupled with a clique-based computation that captures interference of links within the path. We use the topology of a city-wide mesh network deployed in Chaska, MN to demon- strate the effectiveness of our model in accurately predicting throughput and delay under existing traffic conditions and in estimating available path bandwidth. Armed with the ability to estimate available bandwidth over a single path, we use our model to study the ability of routing protocols’ link-cost metrics to discover high throughput paths. Several mesh routing metrics have been proposed that take into account packet loss probability [7], [8], data transmission rates [4], and multi-channel, multi-radio capabilities [9], [16]. These metrics have been demonstrated to find higher throughput paths than minimum-hop metrics. However, their performance has never been investigated under congested conditions that naturally arise in gateway-centric mesh networks. To address this question, we use our model and a series of experiments that gradually evolve from single-link to full-scale city-wide topologies. We find that all existing routing metrics are highly sensitive to traffic load and detect congestion through the packet loss probability. However, we show that packet loss does not always provide accurate information and can result in low-throughput routing decisions, either by selecting congested paths or by filtering-out non-congested paths. Instead, the busy time fraction is an additional factor essential to discover high throughput paths, especially under congested conditions. We introduce a new available bandwidth metric that can be combined with a source-route link-state routing protocol to directly compute the path providing the highest throughput. Our metric takes into account intra-flow and inter-flow inter-