Token-based atomic broadcast using unreliable failure detectors

Abstract

Many atomic broadcast algorithms have been published in the last twenty years. Token-based algorithms represent a large class of these algorithms. Interestingly, all the token-based atomic broadcast algorithms rely on a group membership service, i.e., none of them uses unreliable failure detectors directly. The paper presents the first token-based atomic broadcast algorithm that uses an unreliable failure detector -the new failure detector denoted by R -instead of a group membership service. The failure detector R is compared with ♦P and ♦S. In order to make it easier to understand the atomic broadcast algorithm, the paper derives the atomic broadcast algorithm from a token-based consensus algorithm that also uses the failure detector R.

Easily Rendering Token-Ring Algorithms of Distributed and Parallel Applications Fault Tolerant

“…A work close to ours is R. Ekwall et al's [21]: nodes are organized in a logical ring and the token is sent to the f + 1 successors, where f is the maximum number of failures. However, the goal of the authors' approach differs from ours.…”

A Fault-Tolerant Token-Based Atomic Broadcast Algorithm

“…We want a single token to "glue" the various instances of consensus together. A variation of the algorithm that follows is presented in [19]. The algorithm presented here is easier to understand, with processes that send regular messages (at line 10) and tokens (versus only tokens in [19]).…”

“…The performance of the token based atomic broadcast algorithm presented above has been evaluated in simulation in [19]. This simulation showed that the performance of the new algorithm is better than previous algorithms using failure detectors.…”

“…We consider both local and wide area networks, with several different system sizes and network link characteristics. The new token based algorithm is once again compared to the same two failure detector based algorithms as in [19].…”

“…Furthermore, we consider the two consensus implementations that were already used in the simulated performance study in [19]. Both algorithms use an unreliable failure detector 3S to solve consensus and require at least a majority of correct processes to reach a decision.…”