Quality-aware routing metrics in wireless mesh networks

Abstract

In this chapter we address the problem of selecting good paths in networks made up of multiple wireless links1, such as wireless mesh networks. By "good paths", we mean paths that both benefit individual data transfers (in terms of TCP connection throughput, for example), and which lead to high aggregate network capacity. Finding good paths between nodes in a wireless network involves two steps: 1. Assigning cost metrics to links and paths. 2. Disseminating routing information. The second step, route dissemination, has received much attention over the past decade. The link and/or path metrics need to be disseminated to the nodes in the network using a routing protocol, to help nodes select best paths in a distributed fashion. There are two types of protocols in how the route dissemination is done: proactive and reactive protocols. Proactive protocols determine paths before there is any demand for communication. They calculate the routing tables ahead of time and maintain them through periodic update messages. Examples include Destination-Sequenced Distance Vector Routing (DSDV, [1]), Fisheye State Routing (FSR, [2]), and Optimized Link State Routing (OLSR, [3]). Reactive protocols, on the other hand, do not calculate routes ahead of time. Route discovery follows the communication request. Examples of reactive protocols include Ad Hoc On Demand Distance Vector (AODV, [4]), Temporarily Ordered Routing Algorithm (TORA, [5]) and Dynamic Source Routing (DSR, [6]). In this chapter we address the first issue, assigning cost metrics to links. Regardless of whether a protocol is proactive or reactive, it requires a mechanism to differentiate between different paths. This differentiation is done using cost metrics. The cost metric of a link is the cost of forwarding a packet along the link. The problem of defining a cost metric is considerably harder in wireless networks than in traditional wired networks, because the notion of a "link" between nodes is not well-defined. The properties of the radio channel between any pair of nodes vary with time, and the reliable radio communication range is often unpredictable. The communication quality of a radio channel depends on background noise, obstacles and channel fading, as well as on other transmissions occurring simultaneously in the network. The appropriate cost metric must take into account a number of factors due to the vagaries of radio channels, which in turn makes the task of assigning metrics non-trivial. Moreover, it is desirable that the metrics for the links along a path be composable, so that the end-to-end cost of a path can be easily derived from the individual metrics of the links along the path. We observe that the type of quality aware routing metric to be chosen depends on the physical layer being used. Designing and implementing a physical layer that can fully "hide" the vagaries of the radio channel from higher layers has proven to be difficult for a number of reasons. It requires the physical layer to be able to accurately estimate and adapt several parameters (e.g., transmit power, modulation, error control coding, etc.) to cope with channel conditions that vary rapidly in time. In fact, we are not aware of any current or next-generation radios that propose to employ sophisticated techniques to fully handle channel quality issues at the physical layer, because of implementation complexity and the absence of practically useful codes that can perform well (especially in the non-asymptotic limit of finite packet sizes) across the large range of channel conditions that are observed in practice. Indeed, practical wireless radios such as the ones based on the various IEEE 802 standards (e.g., 802.11, 802.15, etc.) employ only a simple coding strategy, mostly for error detection. Nodes transmit at one of a discrete set of power levels, and rely on a small number of link-layer packet retransmissions to overcome errors. All other packet losses are visible to higher layers, where they may be recovered using end-toend mechanisms (such as TCP retransmissions or packet-level forward error correction implemented by applications). Most wireless mesh networks are radio networks comprised of radios similar to 802.11. Another way modern radios (e.g., 802.11 chip-sets) cope with channel variations is the use of adaptive modulation schemes, allowing higher layers to set one of several possible bit rates. If frame loss rates at a particular bit rate rise, reducing the bit rate can reduce the observed frame loss ratio and improve throughput. Several bit rate adaptation schemes have been proposed (see [7] for a detailed treatment), and the topic remains an active area of work.We view bit rate selection as being complementary to quality-aware routing, in the sense that once the routing protocol picks the best neighbor to use for a destination using measured cost metrics, the link layer picks the best bit rate (modulation scheme) to use for that neighbor. In the presence of bit rate adaptation, some routing metrics may need to be readjusted, and properly normalized with respect to the transmission rate. For instance, for many applications, a packet loss rate of 10% at 10 Mbps may be preferable over a 5% loss rate at 1 Mbps. Hence, a metric based solely on the loss rate should be modified to take the variety of available rates into account. © 2007 Springer Science+Business Media, LLC. All rights reserved.