View in EDS

Send/Receive Patterns Versus Read/Write Patterns in Crash-Prone Asynchronous Distributed Systems

Saved in:
Bibliographic Details
Title: Send/Receive Patterns Versus Read/Write Patterns in Crash-Prone Asynchronous Distributed Systems
Authors: Déprés, Mathilde, Mostéfaoui, Achour, Perrin, Matthieu, Raynal, Michel
Contributors: Taïani, François, Mathilde Déprés and Achour Mostéfaoui and Matthieu Perrin and Michel Raynal
Publisher Information: Array, 2023.
Publication Year: 2023
Subject Terms: Consensus, Asynchrony, Computer systems organization → Fault-tolerant network topologies, Characterization, Test&, Lattice agreement, [INFO] Computer Science [cs], Test&Set, Message-passing, Read/Write register, Distributed Computability, Broadcast abstraction, Networks → Programming interfaces, Computability, Two-process communication, Message pattern, Distributed software engineering, Crash failure, Simplicity, Mutual exclusion, Theory of computation → Distributed computing models, ddc:004, Atomicity, Quorum, Read/write pattern, Set, amp
Description: This paper is on the power and computability limits of messages patterns in crash-prone asynchronous message-passing systems. It proposes and investigates three basic messages patterns (encountered in all these systems) each involving two processes, and compares them to their Read/Write counterparts. It is first shown that one of these patterns has no Read/Write counterpart. The paper proposes then a new one-to-all broadcast abstraction, denoted Mutual Broadcast (in short MBroadcast), whose implementation relies on two of the previous messages patterns. This abstraction provides each pair of processes with the following property (called mutual ordering): for any pair of processes p and p', if p broadcasts a message m and p' broadcasts a message m', it is not possible for p to deliver first (its message) m and then m' while p' delivers first (its message) m' and then m. It is shown that MBroadcast and atomic Read/Write registers have the same computability power (independently of the number of crashes). Finally, in addition to its theoretical contribution, the practical interest of MBroadcast is illustrated by its (very simple) use to solve basic upper level coordination problems such as mutual exclusion and consensus. Last but not least, looking for simplicity was also a target of this article.
Document Type: Conference object
Article
File Description: application/pdf
Language: English
DOI: 10.4230/lipics.disc.2023.16
Access URL: https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.DISC.2023.16
Rights: CC BY
Accession Number: edsair.dedup.wf.002..c984f55b51e3a04aa6e4fa62a73678eb
Database: OpenAIRE
Description
Abstract:This paper is on the power and computability limits of messages patterns in crash-prone asynchronous message-passing systems. It proposes and investigates three basic messages patterns (encountered in all these systems) each involving two processes, and compares them to their Read/Write counterparts. It is first shown that one of these patterns has no Read/Write counterpart. The paper proposes then a new one-to-all broadcast abstraction, denoted Mutual Broadcast (in short MBroadcast), whose implementation relies on two of the previous messages patterns. This abstraction provides each pair of processes with the following property (called mutual ordering): for any pair of processes p and p', if p broadcasts a message m and p' broadcasts a message m', it is not possible for p to deliver first (its message) m and then m' while p' delivers first (its message) m' and then m. It is shown that MBroadcast and atomic Read/Write registers have the same computability power (independently of the number of crashes). Finally, in addition to its theoretical contribution, the practical interest of MBroadcast is illustrated by its (very simple) use to solve basic upper level coordination problems such as mutual exclusion and consensus. Last but not least, looking for simplicity was also a target of this article.
DOI:10.4230/lipics.disc.2023.16