Learning a ring cheaply and fast

We consider the task of learning a ring in a distributed way: each node of an unknown ring has to construct a labeled map of it. Nodes are equipped with unique labels. Communication proceeds in synchronous rounds. In every round every node can send arbitrary messages to its neighbors and perform arbitrary local computations. We study tradeoffs between the time (number of rounds) and the cost (number of messages) of completing this task in a deterministic way: for a given time T we seek bounds on the smallest number of messages needed for learning the ring in time T. Our bounds depend on the diameter D of the ring and on the delay θ = T - D above the least possible time D in which this task can be performed. We prove a lower bound Ω(D2/θ) on the number of messages used by any algorithm with delay θ, and we design a class of algorithms that give an almost matching upper bound: for any positive constant 0 < ε < 1 there is an algorithm working with delay θ ≤ D and using O(D2 (log* D)/θ1 - ε) messages. © 2013 Springer-Verlag