24
0

How Bandwidth Affects the CONGESTCONGEST Model

Abstract

The CONGESTCONGEST model for distributed network computing is well suited for analyzing the impact of limiting the throughput of a network on its capacity to solve tasks efficiently. For many "global" problems there exists a lower bound of Ω(D+n/B)\Omega(D + \sqrt{n/B}), where BB is the amount of bits that can be exchanged between two nodes in one round of communication, nn is the number of nodes and DD is the diameter of the graph. Typically, upper bounds are given only for the case B=O(logn)B=O(\log n), or for the case B=+B = +\infty. For B=O(logn)B=O(\log n), the Minimum Spanning Tree (MST) construction problem can be solved in O(D+nlogn)O(D + \sqrt{n}\log^* n) rounds, and the Single Source Shortest Path (SSSP) problem can be (1+ϵ)(1+\epsilon)-approximated in O~(ϵO(1)(D+n))\widetilde{O}(\epsilon^{-O(1)} (D+\sqrt{n}) ) rounds. We extend these results by providing algorithms with a complexity parametric on BB. We show that, for any B=Ω(logn)B=\Omega(\log n), there exists an algorithm that constructs a MST in O~(D+n/B)\widetilde{O}(D + \sqrt{n/B}) rounds, and an algorithm that (1+ϵ)(1+\epsilon)-approximate the SSSP problem in O~(ϵO(1)(D+n/B))\widetilde{O}(\epsilon^{-O(1)} (D+\sqrt{n/B}) ) rounds. We also show that there exist problems that are bandwidth insensitive.

View on arXiv
Comments on this paper