Design and analysis of adaptive strategies for locating internet-based servers in MANETs

Abstract

A critical problem in providing Internet access to Mobile Ad Hoc Networks (MANETs) is how the mobile hosts can locate Internet-based servers efficiently in a dynamic, unstructured network. This paper studies adaptive strategies that combine both proactive advertising by the servers and on-demand discovery by the mobile hosts. The adaptive strategies determine the relative rate of proactive advertising and on-demand discovery according to system characteristics such as host mobility level and offered load. Our simulation study reveals that, compared to the previous proactive strategies and reactive strategies, the adaptive strategies reduce network traffic by several times when the network has a moderate offered load and a low or high level of host mobility.

Introduction

In a Mobile Ad hoc Network (MANET), a collection of wireless mobile hosts form a temporary network without the assistance of any established infrastructure. A pair of hosts may communicate with each other via other intermediate hosts. Many applications in both the commercial domain and the military domain can be supported by these networks, such as content distribution, information dissemination, ad hoc multimedia streaming, and distributed games, to name just a few.

The approach of combining MANETs with the Internet infrastructure has recently attracted attention [7]. This approach can greatly enrich the applications, increase network scalability, and improve quality of service. First, the Internet supports a much larger set of popular applications. These applications, if made available via the interfaces between the Internet and MANETs, will further facilitate universal accessibility. Second, MANETs are infrastructure-less and they do not scale up to a large number of hosts. For example in large networks, routing and target discovery may incur extremely high traffic. On the other hand, the Internet has a well-established infrastructure. This infrastructure has excellent scalability and abundant network resources. An Internet-based MANET has a more hierarchical architecture: at the high level the Internet connects a set of interfaces between the Internet and MANETs, and at the low level a native MANET connects the mobile hosts. Such a hierarchical approach leads to more scalable systems. Third, in MANETs the network topology changes dynamically and the multi-hop communication route between a pair of hosts can be broken due to host mobility and possibly host joining/leaving. Therefore, it is difficult to provide high-quality service to the applications. On the other hand, by connecting MANETs to the Internet, they will likely provide more reliable services to the applications. If the MANETs are connected to the Internet via many interfaces, a pair of hosts can communicate with each other via an Internet shortcut.

An important question is how to locate the interfaces between the Internet and MANETs. Hereafter, we call the interfaces Internet-based servers in MANETs. In other words, how can the mobile hosts be informed of the existence of the servers and how can they reach the servers? In a stationary network this problem is trivial. However, in a mobile network, every time when a host requests Internet services, it may have to locate a new server and find a path to this server. Due to the infrastructure-less nature of MANETs, it is expensive to do so. The usual search methods of using query flooding generate potentially very high network traffic. Making things worse, when the request rate increases, naive search methods require that search traffic be increased in proportion to the request rate, and hence the network would be overloaded.

This paper focuses on strategies for locating the servers at minimum network cost. Broadly there are two categories of strategies: reactive (on-demand) strategies and proactive strategies. With reactive strategies, a mobile host initiates server discovery only when there is a request. The host floods a query into the network, hoping to hit at least one of the servers and to discover the path to the server. With proactive strategies, the servers advertise their availability by flooding packets into the entire network. In this way the mobile hosts obtain the routes to the servers. There are tradeoffs between the two categories of strategies. A number of factors play important roles in these tradeoffs. At a high level of host mobility, reactive strategies are favored since routes to servers will frequently become stale. At a high offered load (or request rate), proactive strategies will likely cause lower network traffic than reactive strategies do.

We study adaptive strategies for locating the servers in MANETs. By adaptive strategies, we mean, first they are hybrid strategies that combine both proactive advertising by the servers and on-demand discovery by the mobile hosts, and second the relative rate of proactive advertising and on-demand discovery is determined by the current network characteristics (such as request rate and mobility level). We first study the theoretical optimality of adaptive strategies. Our main contributions are two novel, integrated algorithms. First, to determine the rate of proactive advertising, we propose an exponential backoff algorithm to probe the optimal operating point. Second, to reduce the network traffic due to reactive discovery, we propose a novel controlled flooding algorithm, which generates low network traffic. Our simulation study reveals that the combined use of these two algorithms is near-optimal in minimizing network traffic. Compared to the previous proactive strategies and reactive strategies, our adaptive strategies reduce network traffic by several times when the network has a low or high mobility level and a moderate request rate.

The remainder of this paper is organized as follows. In the next section, we first describe our network model, and discuss proactive strategies and reactive strategies. Section 3 illustrates the intuition behind our adaptive strategies and investigates the properties of the optimal operating point. Section 4 describes our exponential backoff algorithm for proactive advertising, and Section 5 describes our novel flooding algorithm. Section 6 describes our simulation and performance evaluation. Section 7 briefly describes related work and finally Section 8 summarizes the paper.