Local computations on static and dynamic graphs

Abstract

The purpose of this paper is a study of computation that can be done locally in a distributed network. By locally we mean within time (or distance) independent of the size of the network. In particular we are interested in algorithms that ore robust, i.e., perform well even if the underlying graph is not stable and links continuously fail and come-up. We introduce and study the happy coloring and orientation problem and show that it yields a robust local solution to the (d,m)-dining philosophers problem of Naor and Stockmeyer [17]. This problem is similar to the usual dining philosophers problem, except that each philosopher has access to d forks but needs only m of them to eat. We give a robust local solution if mles/[d/2] (necessity of this inequality for any local solution was known previously). Two other problems we investigate are: (1) the amount of initial symmetry-breaking needed to solve certain problems locally (for example, our algorithms need considerably less symmetry-breaking than having a unique ID on each node), and (2) the single-step color reduction problem: given a coloring with c colors of the nodes of a graph, what is the smallest number of colors c' such that every node can recolor itself with one of c' colors as a function of its immediate neighborhood only.