Agreement in Partitioned Dynamic Networks

TL;DR

This paper investigates the $k$-agreement problem in $p$-partitioned dynamic networks, providing impossibility results when process count is unknown and algorithms with specific round complexities when an upper bound is known.

Contribution

It introduces new algorithms for $k$-agreement in $p$-partitioned dynamic networks and establishes impossibility results for unknown process counts.

Findings

Impossible to achieve $k$-agreement when process count is unknown for $p extgreater=2$.

Algorithms achieve $k$-agreement in $p(n-p)$ rounds for $k=p$.

Algorithms achieve $O(n/ extepsilon)$ rounds for $k= extlceil (1+ extepsilon)p ceil$.

Abstract

In the dynamic network model, the communication graph is assumed to be connected in every round but is otherwise arbitrary. We consider the related setting of $p$-partitioned dynamic networks, in which the communication graph in each round consists of at most $p$ connected components. We explore the problem of $k$-agreement in this model for $k\geq p$. We show that if the number of processes is unknown then it is impossible to achieve $k$-agreement for any $k$ and any $p\geq 2$. Given an upper bound $n$ on the number of processes, we provide algorithms achieving $k$-agreement in $p(n-p)$ rounds for $k=p$ and in $O(n/\epsilon)$ rounds for $k=\lceil (1+\epsilon)p \rceil$.