Adapting to Asynchronous Dynamic Networks

The computational power of different communication models is a fundamental question in the theory of distributed computation. For example, in the synchronous model messages are assumed to be delivered within one time unit, whereas in the asynchronous model message delays may be arbitrary. Another important parameter of the model is the assumptions about the topology. In the dynamic topology model, links are assumed to crash and recover dynamically, but their status is known to the incident node processors. A meaningful computation can be carried out if the topology stabilizes for a sufficiently long period.

In this paper we show that the model of asynchronous, dynamic-topology network is equivalent, up to polylogarithmic factors, to the synchronous, static (i.e., fixed-topology) model. Specifically, we present a simulation methodology of synchronous static protocols that can withstand arbitrary link delays and changing topology at the expense of only polylogarithmic blowup in the running time, the number of messages, and the space requirement. Previous methods entailed a linear blowup in at least one of these resources.

The generality of our method is demonstrated by a series of improvements for important applications, including Breadth First Search, computing compact efficient routing tables, and packet routing on asynchronous networks.

Click here