High-throughput multicast routing metrics in wireless mesh networks

Abstract

The stationary nature of nodes in a mesh network has shifted the main design goal of routing protocols from maintaining connectivity between source and destination nodes to finding high-throughput paths between them. Numerous link-quality-based routing metrics have been proposed for choosing high-throughput routing paths in recent years. In this paper, we study routing metrics for high-throughput tree or mesh construction in multicast protocols. We show that there is a fundamental difference between unicast and multicast routing in how data packets are transmitted at the link layer, and accordingly how the routing metrics for unicast routing should be adapted for high-throughput multicast routing. We propose a low-overhead adaptive online algorithm to incorporate link-quality metrics to a representative multicast routing protocol. We then study the performance improvement achieved by using different link-quality-based routing metrics via extensive simulation and experiments on a mesh-network testbed, using ODMRP as a representative multicast protocol.

Our extensive simulation studies show that: (1) ODMRP equipped with any of the link-quality-based routing metrics can achieve higher throughput than the original ODMRP. In particular, under a tree topology, on average, ODMRP enhanced with link-quality routing metrics achieve up to 34% higher throughput than the original ODMRP under low multicast sending rate; (2) the improvement reduces to 21% under high multicast sending rate due to higher interference experienced by the data packets from the probe packets; (3) heavily penalizing lossy links is an effective way in the link-quality metric design to avoid low-throughput paths; and (4) the path redundancy from a mesh data dissemination topology in mesh-based multicast protocols provides another degree of robustness to link characteristics and reduces the additional throughput gain achieved by using link-quality-based routing metrics. Finally, our experiments on an eight-node testbed show that on average, ODMRP using SPP and PP achieves 14% and 17% higher throughput over ODMRP, respectively, validating the simulation results.

Introduction

Recently, wireless mesh networks have attracted much attention [45], [46], [47], [48], [49]. These networks typically have low maintenance overhead, high data rates, and are not energy constrained. Such networks, also known as “community wireless networks”, can be used for various applications such as shared broadband access, neighborhood gaming, video surveillance, and media repository.

Unlike traditional mobile ad hoc networks (MANETs), the routers in mesh networks are static, and thus dynamic topology changes are much less of a concern in such networks. As a consequence, the main design goal for routing protocols is shifted from maintaining connectivity between source and destination nodes to finding high-throughput paths between the nodes. Towards this goal, more sophisticated routing metrics than the hop-count metric need to be used to find paths that achieve high throughput, as protocols based on the hop-count metric often choose long links which tend to be lossy and give low throughput.

Unicast is a fundamental routing service in multihop mesh networks. As such, much work has been done recently on finding a high-throughput routing metric for unicast routing protocols in wireless mesh networks. For example, the expected transmission count (ETX) metric was proposed in [8] to take into account the loss rates of the links. Similarly, the round trip time (RTT) and packet pair (PP) metrics were proposed in [1], [22], respectively, to take into account the delay characteristics of the links. Finally, the Weighted cumulative expected transmission time (WCETT) metric was proposed in [13] to take into account the link bandwidth and loss rates of links in the presence of multiple non-overlapping channels. In [4], [11], two link-quality-based routing metrics were proposed to minimize the expected amount of energy needed for end-to-end data transmission assuming a reliable link layer and an unreliable link layer, respectively. All these metrics have been proposed and evaluated for unicast routing protocols such as DSDV [35], DSR [21], and AODV [36].

Multicast is another fundamental routing service in multihop mesh networks. It provides an efficient means of supporting collaborative applications such as video conferencing, online games, webcast and distance learning, among a group of users [10]. Previously, a large number of multicast protocols have been designed to efficiently maintain a distributed multicast routing structure in dynamically changing topologies in MANETs [38], [18], [26], [43], [16], [39], [19], [17], [31]. All of these protocols use minimum-hop-count as the routing metric and focus on scenarios with high mobility. Similarly as for unicast protocols, the stationary nature of nodes in a mesh network also shifts the main design goal of multicast routing protocols from efficiently maintaining a distributed multicast routing structure to finding a high-throughput multicast routing structure such as a tree or a mesh. Despite the numerous recent studies on link-quality-based routing metrics for high-throughput unicast routing in mesh networks, there has been little work on link-quality-based routing metrics for high-throughput multicast routing.

In this paper, we study the design of link-quality-based routing metrics for high-throughput multicast in mesh networks. One approach is to simply adopt the unicast routing metrics such as ETX and WCETT for multicast protocols, i.e., in the tree or mesh construction. However, these metrics may not work well as the way they reflect the link quality depends very much on the link layer data transmission, and multicast routing is inherently different from unicast routing in the way the link layer handles data packets. In particular, while unicast routing protocols use unicast to send data packets at the link layer, most multicast routing protocols use broadcast at the link layer for disseminating data packets. The fundamental difference between link layer broadcast and link layer unicast is that the former has no link layer acknowledgments and consequently no link layer retransmissions. This difference has two immediate implications on multicast routing: (1) the link quality that matters is unidirectional, since per-hop data transmission no longer involves two round trips (RTS/CTS/DATA/ACK); and (2) each node has only one chance to properly transmit a data packet at the link layer since there are no retransmissions. These implications suggest that link-quality metrics designed for unicast may no longer be appropriate for multicast, i.e., for constructing a high-throughput multicast tree or mesh.

In this paper, we first study how to adapt the routing metrics developed for unicast for use in multicast in mesh networks. We then study the performance of a set of five routing metrics adapted from those for unicast protocols, namely, ETT, ETX, PP, Multicast ETX (METX) and Success Probability Product (SPP), where METX and SPP are adapted from two energy-efficient routing metrics proposed in [4], [11]. Our study is performed using ODMRP [26], a state-of-the-art proactive multicast protocol. For each routing metric, we modify ODMRP to construct the routing structure based on that routing metric. We also compare the modified versions of ODMRP with the original ODMRP. Our studies are conducted via extensive simulations as well as experiments on a mesh network testbed consisting of eight nodes deployed on the second floor of an office building on the Purdue campus.

Our simulation study shows that ODMRP using any of the five metrics, ETT, ETX, METX, PP, and SPP, outperforms the original ODMRP by significant margins. These margins of improvement are similar to those achieved in unicast routing using high-throughput routing metrics [12]. In particular, on average, ODMRP using SPP or PP achieves 34% and 21% higher throughput than the original ODMRP under low and high load, respectively. Our testbed experiments show that on average, ODMRP using SPP and PP achieves 14% and 17% higher throughput over ODMRP, respectively. Finally, our simulation studies show that path redundancy in a mesh topology provides another degree of robustness to link characteristics. In particular, when the number of sources per group is increased from one to two, the increased path redundancy in the original ODMRP causes it to perform reasonably well, and the additional throughput improvement from using link-quality-based routing metrics is reduced by around 15% under both low and high loads.

The main contributions of this paper are as follows:

• We point out that high-throughput routing metrics design for unicast routing cannot be directly applied to multicast routing due to a fundamental difference between unicast and multicast routing in how data packets are transmitted at the link layer.

• We show how to adapt the routing metrics for unicast routing for high-throughput multicast routing.

• We propose a low-overhead online adaptive scheme for incorporating link-quality metrics to a representative multicast protocol.

• We present detailed simulation results of ODMRP based on different routing metrics which demonstrate significant throughput gain from using high-throughput routing metrics.

• We present experimental results on a testbed that validate the simulation results.

• We present tradeoffs between probing overhead and throughput gain in using high-throughput routing metrics.

• We show that path redundancy in a mesh topology provides another degree of robustness to link characteristics which reduces the throughput gain from using high-throughput routing metrics.

To the best of our knowledge, this is the first study on high-throughput routing metrics for multicast in wireless mesh networks.

The rest of the paper is organized as follows. Section 2 describes the routing metrics previously proposed for unicast routing protocols. Section 3 discusses the fundamental difference between multicast and unicast modes of communication in wireless multihop networks and describes how to accordingly modify the existing unicast routing metrics for multicast routing. Section 2.2 gives a brief overview of ODMRP and Section 4 the changes made to ODMRP in order to incorporate the routing metrics. Section 5 presents simulation results and Section 6 presents experimental results on a mesh network testbed. Finally, Section 7 reviews related work and Section 8 concludes the paper.