Scheduling for shared window joins over data stream

Moustafa A. Hammad, Michael J. Franklin, Walid G. Aref, Ahmed K. Elmagarmid

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

110 Citations (Scopus)

Abstract

Continuous Query (CQ) systems typically exploit commonality among query expressions to achieve improved efficiency through shared processing. Re cently proposed CQ systems have introduced window specifications in order to support unbounded data streams. There has been, however, little investigation of sharing for windowed query operators. In this paper, we address the shared execution of windowed joins, a core operator for CQ sys tems. We show that the strategy used in systems to date has a previously unreported performance flaw that can negatively impact queries with relatively small windows. We then propose two new execution strategies for shared joins. We evaluate the alternatives using both analytical model and implementation in a DBMS. The results show that one strategy, called MQT, provides the best performance over a range of workload settings.

Original languageEnglish
Title of host publicationProceedings - 29th International Conference on Very Large Data Bases, VLDB 2003
EditorsPatricia G. Selinger, Michael J. Carey, Johann Christoph Freytag, Serge Abiteboul, Peter C. Lockemann, Andreas Heuer
PublisherMorgan Kaufmann
Pages297-308
Number of pages12
ISBN (Electronic)0127224424, 9780127224428
Publication statusPublished - 2003
Externally publishedYes
Event29th International Conference on Very Large Data Bases, VLDB 2003 - Berlin, Germany
Duration: 9 Sept 200312 Sept 2003

Publication series

NameProceedings - 29th International Conference on Very Large Data Bases, VLDB 2003

Conference

Conference29th International Conference on Very Large Data Bases, VLDB 2003
Country/TerritoryGermany
CityBerlin
Period9/09/0312/09/03

Fingerprint

Dive into the research topics of 'Scheduling for shared window joins over data stream'. Together they form a unique fingerprint.

Cite this