Presentation 1997/1/23
Simulation Study of a Runtime Load Balancing Algorithm for Pipelined Hash Multi-Joins
Stephen Davis, Masaru Kitsuregawa,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) The regular quadrupling of the available memory space and the scalability of shared nothing systems now allow for the storage of multiple relations in memory. As a result, pipelined processing of multi-join operations is now feasible. However, if care is not taken to evenly distribute the work load among the processors, performance degrades. Current approaches to dynamic load balancing work to balance the load as a part of data partitioning, and provide no mechanisms for detecting and resolving skew during actual join processing. Because it is very difficult to combine partitioning with load balancing in an efficient manner, it is important that methods of resolving skew in processing load be included in a shared-nothing multi-join operation. In this paper, we present a runtime load balancing algorithm which utilizes a centralized load manager to examine the load distributions during join processing, and to determine how build relation tuples should be redistributed in order to resolve any imbalances.
Keyword(in Japanese) (See Japanese page)
Keyword(in English)
Paper # DE96-75
Date of Issue

Conference Information
Committee DE
Conference Date 1997/1/23(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Data Engineering (DE)
Language ENG
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Simulation Study of a Runtime Load Balancing Algorithm for Pipelined Hash Multi-Joins
Sub Title (in English)
Keyword(1)
1st Author's Name Stephen Davis
1st Author's Affiliation University of Tokyo, Institute of Industrial Science()
2nd Author's Name Masaru Kitsuregawa
2nd Author's Affiliation University of Tokyo, Institute of Industrial Science
Date 1997/1/23
Paper # DE96-75
Volume (vol) vol.96
Number (no) 469
Page pp.pp.-
#Pages 6
Date of Issue