Consensus in the Crash-Recover Model
Rui Oliveira, Rachid Guerraoui, A. Schiper · 1997
This paper presents a deterministic algorithm that solves consensus in asynchronous distributed systems where processes may crash and recover, messages may be lost, and failure detections may be inaccurate. Our algorithm has an early delivery property: in runs where no process crashes or is suspected to have crashed, consensus is reached only after two communication steps. The paper shows that to solve consensus with processes that may crash and recover, more knowledge about failures is required than if we assume that crashed processes never recover. Key words: Distributed algorithms - Asynchronous system - Unreliable failure detector - Crash-recover model - Consensus 1 Introduction