Adaptive wireless mesh networks: Surviving weather without sensing it Nauman Javed, Eric Lyons, Michael Zink, Tilman Wolf ⇑ Department of Electrical and Computer Engineering, University of Massachusetts, Amherst, MA 01003, USA article info Article history: Received 13 December 2013 Received in revised form 2 July 2014 Accepted 7 August 2014 Available online 22 September 2014 Keywords: Routing Self-conﬁguring network Wireless mesh network Prediction Distributed algorithm abstract Large-scale wireless mesh networks, like the ones used as cellular back-haul, operate under circum- stances, where individual links are affected by weather conditions. Reliability requirements in wireless mesh networks necessitate the ability to reconﬁgure the network in the face of changing environmental conditions. In this paper, we present a predictive routing protocol for wireless mesh networks, which operate at millimeter-wave bands with directional links, that uses in-network parameter prediction to make the network adaptive, as opposed to using meteorological weather information from external sources, such as weather radars. We validate our approach through simulations based on real-world weather events, observed through a network of weather radars, and comparisons with approaches that do not make use of predictions but may use the link quality as a parameter in routing decision making. Our results show that our link quality-based predictive approach can achieve throughput performance that is almost 8% better than a link quality-based routing algorithm that does not use prediction for the real weather scenario we use for our simulations. Ó 2014 Elsevier B.V. All rights reserved. 1. Introduction and motivation Many networked systems consist of geographically distributed elements and are affected by geographically distributed environ- mental phenomena, internal or external to the network. This is obviously true for Distributed Sensor Networks [1], which are affected by, and in fact measure, the environmental variables external to the network. These environmental phenomena are part of the physical world where these Distributed Sensor Networks are situated, but not some internal characteristics of the networks. Dis- tributed implementations of the parameter estimation and infer- ence are important for these Distributed Sensor Networks. In this paper, we look at an example of a Distributed Cyber-Physical Sys- tem that does not directly observe, or make inferences about, the parameters of the surrounding physical environment, but is, any- way, affected by them [2]. This network, an example of a Distrib- uted Cyber-Physical System, instead monitors and makes predictive inferences about the parameters within the network that were indeed originally affected by the parameters of the physical environment external to the network. The goal of such in-network predictions is to be able to make network-wide conﬁguration. Our example Distributed Cyber-Physical System is a Wireless Mesh Network for which we aim to design a weather disruption-tolerant, predictive, routing protocol. There are many examples of networked systems that are affected by geographically distributed environmental phenomena, internal or external to the network. Examples include:  The routers and links of the Internet, or other computer net- works, affected by the geographically, or in general logically, distributed trafﬁc load patterns.  The performance of a wireless network affected by geographi- cally distributed weather phenomena through their impact either on the wireless propagation, or other elements of the wireless communication system, such as the antennas. As the performance of the services provided by the networked systems in the above-mentioned examples is affected by some geographically distributed environmental phenomena, methods are required to alleviate the adverse effects on the performance of these systems. For example, routing protocols in networks aim to establish routes between multiple source–destination pairs that are best in some sense, such as throughput, and delay. The alleviation of network performance degradation by distrib- uted protocols, such as network routing protocols, has an inherent lag due to the amount of time it takes to detect the performance degradation and then reconﬁguring the network. In this paper, we propose to develop methods of network conﬁguration based on distributed estimation and prediction of network performance http://dx.doi.org/10.1016/j.comcom.2014.08.007 0140-3664/Ó 2014 Elsevier B.V. All rights reserved. ⇑ Corresponding author. Tel.: +1 413 545 0757. E-mail addresses: njaved@ecs.umass.edu (N. Javed), lyons@ecs.umass.edu (E. Lyons), zink@ecs.umass.edu (M. Zink), wolf@ecs.umass.edu (T. Wolf). Computer Communications 54 (2014) 120–130 Contents lists available at ScienceDirect Computer Communications journal homepage: www.elsevier.com/locate/comcom