Communications of the ACM

In this article, we study the problem of distributed selection from a theoretical point of view. Given a general connected graph of diameter D consisting of n nodes in which each node holds a numeric element, the goal of a k-selection algorithm is to determine the k^th smallest of these elements. We prove that distributed selection indeed requires more work than other aggregation functions such as, e.g., the computation of the average or the maximum of all elements. On the other hand, we show that the k^th smallest element can be computed efficiently by providing both a randomized and a deterministic k-selection algorithm, dispelling the misconception that solving distributed selection through in-network aggregation is infeasible.