Swarm consensus

TL;DR

The paper introduces a leaderless consensus algorithm for gnome swarms based on gossip protocols, achieving reliable collective decision-making with bounded reaction time, and extends it to Byzantine fault tolerance.

Contribution

It presents a novel leaderless consensus algorithm inspired by gnome swarm behavior, including a Byzantine fault-tolerant variant, with analysis of reaction time bounds.

Findings

Consensus achieved within 2d t rounds

Algorithm is non-Byzantine and resilient to failures

Extended to Byzantine fault tolerance

Abstract

The strength of gnomes lies in their coordinated action. Being small and subtle creatures themselves, the forest gnomes can form large swarms acting as one giant creature. This unusual defense strategy requires a lot of skill and training. Directing a swarm is not an easy task! Initially, gnomes used leader-based control algorithms, although those have been proven to be vulnerable to abuse and failure. After thorough research and study, gnomes developed their own leaderless consensus algorithm based on very simple rules. It is based on gossip in a network of a known diameter $d$. One of the gnomes proposes a plan which then spreads gnome to gnome. If there is an agreement, gnomes act \emph{all at once}. If there are conflicting plans (an extreme rarity), they try again. The resulting upper bound on the swarm's reaction time is its round-trip time $2dt$, where $t$ is the command relay time. The original algorithm is non-Byzantine; all gnomes must be sane and sober. While working on the algorithm, gnomes discovered \emph{swarm time}, a sibling concept to L.Lamport's logical time. That led to a Byzantine-ready version of the algorithm.