Reducing Congestion Effects in Wireless Networks by Multipath Routing Lucian Popa 1 , Costin Raiciu 2 , Ion Stoica 1 , David S. Rosenblum 2 1 Department of Computer Science University of California, Berkeley {popa, istoica}@cs.berkeley.edu 2 Department of Computer Science University College London {c.raiciu, d.rosenblum}@cs.ucl.ac.uk Abstract—We propose a solution to improve fairness and in- crease throughput in wireless networks with location informa- tion. Our approach consists of a multipath routing protocol, Bi- ased Geographical Routing (BGR), and two congestion control algorithms, In-Network Packet Scatter (IPS) and End-to-End Packet Scatter (EPS), which leverage BGR to avoid the con- gested areas of the network. BGR achieves good performance while incurring a communication overhead of just 1 byte per data packet, and has a computational complexity similar to greedy geographic routing. IPS alleviates transient congestion by splitting traffic immediately before the congested areas. In con- trast, EPS alleviates long term congestion by splitting the flow at the source, and performing rate control. EPS selects the paths dynamically, and uses a less aggressive congestion control mechanism on non-greedy paths to improve energy efficiency. Simulation and experimental results show that our solution achieves its objectives. Extensive ns-2 simulations show that our solution improves both fairness and throughput as compared to single path greedy routing. Our solution reduces the variance of throughput across all flows by 35%, reduction which is mainly achieved by increasing throughput of long-range flows with around 70%. Furthermore, overall network throughput in- creases by approximately 10%. Experimental results on a 50- node testbed are consistent with our simulation results, suggest- ing that BGR is effective in practice. I. INTRODUCTION Wireless embedded processors contained in mobile phones, handheld devices or weaved into the environment as sensors, are likely to become the main part of the future Internet [9]. Furthermore, it is expected that location information will be widely available for such processing, to enhance context-aware types of interactions [9]. The prospect of having ad-hoc wireless networks com- posed of numerous location-aware nodes spread in the sur- rounding environment (such as SmartDust [1]) poses new in- teresting challenges to the research community. Congestion in wireless networks has already been explored by other research, observing its impact on performance: a drastic decrease in throughput [28] and increased per-packet energy consumption [11]. On the other hand, computing is moving to an era where applications require large and stable bandwidths to perform their tasks. Such applications include multimedia applications, high frequency sensing applications, file transfer, and so forth. If devices enabling these applications are going to become an integral part of tomorrow’s networks, solutions to reduce the effects of congestion in wireless networks are required. A promising approach for routing in such networks is geo- graphical routing, an algorithm that leverages location infor- mation to route messages in a hop-by-hop, greedy manner. Assuming that a coordinate system is in place (either GPS or other coordinate systems, such as NoGeo [2], BVR [4] or [3]), this scheme is scalable, has low computational overhead and requires minimum routing information to be maintained by nodes. However, shortest path routing schemes in general, and geographical routing in particular, amplify the effects of con- gestion: in a random communication pattern, the nodes in the center of the network carry a disproportionately large amount of the entire traffic, drastically decreasing the throughput of the flows they forward. This affects most long-range flows, as they have a higher probability of intersecting the central hot- spot. In this paper, we present a solution that seeks to utilize idle or under-loaded nodes to reduce the effects of congestion. To achieve this goal, we enhance geographic routing to allow a source to select different paths towards the destination. While multi-path solutions for geographic routing have been pro- posed before, they have either limited effectiveness (e.g., way- point routing), or they exhibit a high overhead (e.g., TBF [5]). At this end, we propose Biased Geographical Routing (BGR), a lightweight, stateless, geographical forwarding algorithm, as a cost-effective complement to greedy routing. BGR routes packets on curved trajectories, by forwarding packets along curves, instead of along the shortest path, towards the destina- tion. To further mitigate congestion, we design two congestion control mechanisms that leverage BGR: • In-Network Packet Scatter (IPS) is a lightweight mechanism that aims to relieve transient congestion by locally splitting the traffic along multiple paths to avoid congested hotspots. • End-to-End Packet Scatter (EPS) is an end-to-end mecha- nism that aims to alleviate longer term congestion, when IPS fails. EPS works by splitting the flow at the source, and per- forming independent rate control along each path in re- sponse to congestion. We have evaluated the performance of BGR by using a high-level simulator, a packet-level simulator (ns2 [6]), and a testbed comprising 90 nodes [21]. The results show that BGR is a practical and efficient multipath routing algorithm. We have evaluated IPS and EPS using ns2. Simulation results show that their combined action: • increases network throughput for long flows with around 70% when compared to greedy routing, • increases fairness by reducing the dependence of flow throughput on the distance between the endpoints, and • increases overall network throughput by around 10%. In addition, we have evaluated the potential of multipath routing to increase the network throughput on the Mirage test- bed [21]. Experimental results are consistent with the simula- 96 1-4244-0593-9/06/$20.00 ©2006 IEEE