Impact of caching and MAC overheads on routing performance in ad hoc networks

Abstract

We consider an on-demand routing protocol for mobile ad hoc networks that uses route caching aggressively. We show that stale routes in the caches and medium access control (MAC) overhead for replies from caches can degrade performance significantly, so much so that relative performance is much better without using replies from cache. We analyze these problems in a series of evaluation steps, usually looking at both routing and MAC layer performances together. We suggest several mechanisms to address the stale cache problem and also suggest a change in the MAC layer interface to reduce the MAC overhead. The combination of these techniques improves the relative performance of cached replies. Apart from improving performance of the routing protocol, this work serves as a case study where an analysis spanning multiple layers was necessary to identify the performance bottlenecks in a wireless network.

Introduction

In an ad hoc network, mobile nodes communicate with each other using multi-hop wireless links. There is no stationary infrastructure such as base stations or access points. Each node in the network also acts as a router, forwarding data packets for other nodes. Study of ad hoc networks has recently attracted significant interest, for primarily two reasons. First, the very idea of automatically configurable, structure-less network, able to operate in a very dynamic environment with a high degree of fault-tolerance is interesting, and lot of future networked systems can utilize such properties. Examples range from military networks to sensor networks. Second, design of such networks presents interesting technical challenges. The challenges not only span across different layers of the traditional protocol stack, but also involve complex cross-layer interactions. While progress has been made in designing and evaluating protocols in the network [1], [2], [3], [4], [5] and medium access control (MAC) layers [6], [7], [8], [9] in isolation, relatively little attention has been paid in studying cross-layer interactions. In this paper, our thesis is that a holistic view of all protocol layers is important for the best overall performance. We present a case study illustrating this point with a step-by-step performance evaluation and improvement process that looks at both the network and MAC protocols together.

Ad hoc networks present two fresh challenges in network design. First, in the network layer, a dynamic routing protocol must efficiently find routes between two communicating nodes. The routing protocol must be able to keep up with the high degree of node mobility that often changes the network topology drastically and unpredictably. Second, in the link layer, a MAC protocol must provide access to the wireless medium efficiently and reduce interference. The impact of routing overhead and MAC control overheads or transmission errors can be significant in bandwidth-poor wireless networks. They can take a major toll on the overall performance unless controlled carefully.

In the second half of the past decade, there has been a considerable activity in designing protocols for both these layers in the ad hoc or packet radio framework. In particular, a new class of dynamic routing protocols, called ‘on-demand’ protocols [1], [2], has been invented that discovers routes on-demand, rather than proactively as in the distributed shortest-path protocols used in the Internet. The on-demand protocols are able to keep the routing overheads low without compromising significantly on the quality (cost) of the routes discovered [3], [4], [10]. They do this by eliminating unnecessary route finding activities and by using route caching directly or indirectly. For ease of scalability, combinations of proactive and on-demand routing have also been studied [11]. In the MAC layer, several variations of traditional carrier-sense multiple access (CSMA) techniques have been investigated to skirt around the well-known hidden-terminal and exposed terminal problems [12]. Important examples include CSMA with collision avoidance that uses a random back-off even after the carrier is sensed idle [9]; and a virtual carrier sensing mechanism using request-to-send/clear-to-send (RTS/CTS) control packets [6]. Both techniques are used in the IEEE wireless LAN standard 802.11 [9]. In addition to network and link layers, performance of transport layer protocols like TCP has also been investigated [13], [14].

What is missing, however, from the extant literature is a sense of the big picture—a wholesome view of the dynamics of multiple protocol layers that tell a coherent story of what impacts network performance. In the following sections we choose one well-studied routing protocol (dynamic source routing or DSR [1], [15]) and show how a subtle variation of this protocol can tremendously and somewhat unexpectedly improve performance. We investigate the reason of this altered performance delving deeper in the protocol stack. We show that the performance in the original protocol is impacted by a certain caching phenomena in the routing layer and MAC control overheads in the link layer, with their complicity taking a high toll on performance. We suggest solutions at both layers to alleviate these problems. Our goal with this performance study is not merely to improve performance of one single routing protocol, but also to present a systematic study demonstrating how interactions across layers could be studied to understand the performance of a wireless network.

The rest of the paper is organized as follows. In Section 2, we briefly describe the DSR protocol with particular emphasis on its caching activities and related performance issues. We also describe the IEEE 802.11 MAC layer protocol that will be used in the case study. Section 3 discusses related work. In Section 4, we describe the simulation environment and performance metrics. In Section 5, we present the complete performance study with different protocol variations in a sequence of steps. This section forms the core of our work. Finally, we conclude in Section 6.