Distributed coordination in mobile wireless environments

Abstract

Advances in wireless networking technologies and powerful portable mobile devices have engendered the new paradigm of mobile computing, whereby mobile users carrying portable devices can access the information and services for various tasks regardless of their physical locations or movement behaviors. Mobile computing is a branch of distributed computing, but mobile networks have fundamentally different characteristics from traditional wired networks in aspects of communication, mobility and resource constraints. These characteristics make the development of distributed algorithms much more difficult. In this thesis, we investigate the challenging issues in designing algorithms for solving distributed computing problems in mobile wireless networks. We focus on two distributed coordination problems: the consensus problem and the mutual exclusion problem. The consensus problem arises in many distributed computing applications, such as atomic commitment, atomic broadcast, and file replication. So far, little work has been reported on achieving consensus in mobile environments. This thesis makes the following original contributions in this field. First, we develop a general technique named "Look-Ahead" to speed up the execution of consensus protocols by making use of future messages. Nearly all existing consensus protocols for asynchronous systems are executed in asynchronous rounds and each round is divided into several phases. Due to the asynchrony, some "future" message may be delivered to a receiver that has not yet entered the phase or round of the message. By making use of such future messages, hosts can decide to stop waiting for a message with a long delay or sent late, so as to speed up their executions. Second, we improve message efficiency and scalability of consensus protocols for mobile ad hoc networks (MANETs) using a hierarchy imposed on the mobile hosts. By clustering the mobile hosts into clusters, a two-layer hierarchy is established. Then, the messages from and to the hosts in the same cluster are merged/unmerged by the clusterhead in order to reduce the message cost and improve the scalability. Based on different ways for clustering hosts, we propose two hierarchical protocols. The first protocol is based on a static set of clusterheads, in which the function of clustering hosts and the function of achieving consensus are interlaced. In the second hierarchical protocol, clusterheads are dynamically selected, and the functions of clustering hosts and of achieving consensus are separated using a modular approach. With such an approach, the design of hierarchical consensus protocols is simplified, similar as the separation of failure detection and achieving consensus in failure detector based protocols. We define a new oracle, named "eventual clusterer", which is in charge of the construction and maintenance of the hierarchy in a MANET. Based on the eventual clusterer oracle, we design a hierarchical consensus protocol. The third contribution to the consensus problem is the design of an eventual leader protocol for "dynamic" infrastructured mobile networks, where the number of participating hosts can change arbitrarily as time passes and an unbounded number of hosts can join or leave the system at any time. The proposed eventual leader protocol can be used to design consensus protocols for dynamic infrastructured mobile networks. Another coordination problem addressed in this thesis is mutual exclusion (MUTEX), one of typical coordination problems, which is concerned with the coordination of accesses to critical section (CS) mutual exclusively. We propose the first permission-based MUTEX algorithm for MANETs. Unlike token-based algorithms, a permission-based algorithm needs neither to maintain any logical topology nor to propagate any message if no host requests the CS. However, a permission-based algorithm has to send messages for each request of hosts, which significantly increases the average message cost. Based on the "look ahead" technique, which enforces the MUTEX only among the hosts that are currently competing for CS, we propose a message efficient MUTEX algorithm for MANETs. The algorithm can also tolerate link and host failures by using timeout-based fault tolerance mechanisms.