Dynamic networks: models and algorithms

The study of dynamic networks has come into popularity recently, and many models and algorithms for such networks have been suggested. In this column we survey some recent work on dynamic network algorithms, focusing on the effect that model parameters such as the type of adversary, the network diameter, and the graph expansion can have on the performance of algorithms. We focus here on high-level models that are not induced by some specific mobility pattern or geographic model (although much work has gone into geographic models of dynamic networks, and we touch upon them briefly in Section 2). Dynamic network behavior has long been studied in distributed computing literature, but initially it was modeled as a fault in the network; as such, it was typically bounded, either in duration or in the number of nodes affected (or both). For example, in the general omission-fault model, if two nodes that could once communicate can no longer send messages to each other, this is treated as a failure of one of the nodes, and the number of faulty nodes is assumed to be bounded. Another example is self-stabilizing algorithms, which are guaranteed to function correctly only when changes to the network have stopped [16]. These models are appropriate for modeling unreliable static networks, but they are not appropriate for mobile and ad hoc networks, where changes are unbounded in number and occur continually. In the sequel we survey several models for dynamic networks, both random and adversarial, and algorithms for these models. The literature on dynamic networks is vast, and this column is not intended as a comprehensive survey. We have chosen to focus on models and algorithms that exhibit the following properties.