An improved selection-based parallel range-join algorithm in hypercubes

Hong Shen

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

4 Citations (Scopus)

Abstract

The range-join of two sets R and S is the set that contains all tuples (r, s) satisfying e1 ≤|r - s| ≤e2, r ε R and s ε S. For computing the range-join of R and S in a hypercube of p processors, this paper presents an improved selection-based parallel algorithm which reduces the local memory from O(n) required in the previous algorithm [10] to Q(m+n/p), where |R| = m, |S| - n and p ≤ max{m, n}. The new algorithm also reduces the best-case time complexity from 0( m/p log2 p+n/p logm) of the previous result to 0(m+n/p log2p) when m ≥ plogP, while maintaining the cost optimality in the worst case. Unlike the previous algorithm, our algorithm works by selecting the median of R U S to evenly partition the whole data set for divide-and-conquer join in the next phase. We present an upper bound of time complexity of the algorithm in the general case and show that the best-case time complexity of the algorithm is better than permutation-based range-join [9] when n ≥ plogp+1.

Original languageEnglish
Title of host publicationProceedings of the 20th EUROMICRO Conference on System Architecture and Integration, EUROMICRO 1994
PublisherIEEE Computer Society
Pages65-72
Number of pages8
ISBN (Print)0818664304, 9780818664304
DOIs
Publication statusPublished - 1994
Externally publishedYes
Event20th EUROMICRO Conference on System Architecture and Integration, EUROMICRO 1994 - Liverpool, United Kingdom
Duration: 8 Sept 19948 Sept 1994

Publication series

NameConference Proceedings of the EUROMICRO
ISSN (Print)1089-6503

Conference

Conference20th EUROMICRO Conference on System Architecture and Integration, EUROMICRO 1994
Country/TerritoryUnited Kingdom
CityLiverpool
Period8/09/948/09/94

Fingerprint

Dive into the research topics of 'An improved selection-based parallel range-join algorithm in hypercubes'. Together they form a unique fingerprint.

Cite this