Staleness and Local Progress in Transactional Memory

Hagit Attiya, Panagiota Fatourou, Sandeep Hans, Eleni Kanellou

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

Abstract

A key goal in the design of Transactional Memory (TM) systems is ensuring liveness. Local progress is a liveness condition which ensures that a process successfully completes every transaction it initiates, if it continually re-invokes it each time it aborts. In order to facilitate this, several state-of-the-art TM systems keep multiple versions of data items. However, this method can lead to high space-related overheads in the TM implementation. Therefore, it is desirable to strike a balance between the progress that a TM can provide and its practicality, while ensuring correctness. A consistency property that limits the number of previous versions a TM may rely on, is k-staleness. It is a condition derivative of snapshot isolation, in which a transaction is not allowed to access more than k previous versions of a data item. This facilitates implementations that can take advantage of multi-versioning, while at the same time, contributing to the restriction of the space overhead introduced by the TM. In this paper, we prove that no TM can ensure both local progress and k-staleness, if it is unaware of the transaction’s accesses and can only keep a bounded number of versions for each data item.

Original languageEnglish
Title of host publicationNetworked Systems - 8th International Conference, NETYS 2020, Proceedings
EditorsChryssis Georgiou, Rupak Majumdar
PublisherSpringer Science and Business Media Deutschland GmbH
Pages227-243
Number of pages17
ISBN (Print)9783030670863
DOIs
StatePublished - 2021
Event8th International Conference on Networked Systems, NETYS 2020 - Marrakech, Morocco
Duration: 3 Jun 20205 Jun 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12129 LNCS

Conference

Conference8th International Conference on Networked Systems, NETYS 2020
Country/TerritoryMorocco
CityMarrakech
Period3/06/205/06/20

Keywords

  • Consistency
  • Impossibility
  • Progress
  • Transactional memory

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Staleness and Local Progress in Transactional Memory'. Together they form a unique fingerprint.

Cite this