Transforming worst-case optimal solutions for simultaneous tasks into all-case optimal solutions

Maurice P. Herlihy, Yoram Moses, Mark R. Tuttle

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Decision tasks require that nonfaulty processes make decisions based on their input values. Simultaneous decision tasks require that nonfaulty processes decide in the same round. Most decision tasks have known worst-case lower bounds. Most also have known worst-case optimal protocols that halt in the number of rounds given by the worst-case lower bound, and some have early-stopping protocols that can halt earlier than the worst-case lower bound (sometimes in as early as two rounds). We consider what might be called earliest-possible protocols for simultaneous decision tasks. We present a new technique that converts worst-case optimal decision protocols into all-case optimal simultaneous decision protocols: For every behavior of the adversary, the all-case optimal protocol decides as soon as any protocol can decide in a run with the same adversarial behavior. Examples to which this can be applied include set consensus, condition-based consensus, renaming and order-preserving renaming. Some of these tasks can be solved significantly faster than the classical simultaneous consensus task. A byproduct of the analysis is a proof that improving on the worst-case bound for any simultaneous task by even a single round is as hard as reaching simultaneous consensus.

Original languageEnglish
Title of host publicationPODC'11 - Proceedings of the 2011 ACM Symposium Principles of Distributed Computing
Pages231-238
Number of pages8
DOIs
StatePublished - 2011
Event30th Annual ACM SIGACT-SIGOPS Symposium on Principles of Distributed Computing, PODC'11, Held as Part of the 5th Federated Computing Research Conference, FCRC - San Jose, CA, United States
Duration: 6 Jun 20118 Jun 2011

Publication series

NameProceedings of the Annual ACM Symposium on Principles of Distributed Computing

Conference

Conference30th Annual ACM SIGACT-SIGOPS Symposium on Principles of Distributed Computing, PODC'11, Held as Part of the 5th Federated Computing Research Conference, FCRC
Country/TerritoryUnited States
CitySan Jose, CA
Period6/06/118/06/11

Keywords

  • common knowledge
  • condition-based consensus
  • consensus
  • crash failure model
  • k-set agreement
  • renaming
  • synchronous message passing model
  • topology

All Science Journal Classification (ASJC) codes

  • Software
  • Hardware and Architecture
  • Computer Networks and Communications

Fingerprint

Dive into the research topics of 'Transforming worst-case optimal solutions for simultaneous tasks into all-case optimal solutions'. Together they form a unique fingerprint.

Cite this