Presentation 1997/9/26
Towards Dynamic Clustering
Jun Kiniwa, Tiko Kameda,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In two-level memories including some fast accessible pages, we investigate an effective clustering of data in each page. Given an offine sequence of requests, we derive the ratio bound of the worst clustering to the optimal clustering if the optimal page replacement policy MIN is used. Next we present two heuristic methods, called Longest Unused and Nearest Neighbor, which can generate desirable clusterings. If we can do clustering beforehand by using our methods, the bound above can be highly improved in the Nearest Neighbor, in particular. Assuming the locality of reference, we also examine the performance of our methods by simulation for online sequences of requests and the page replacement policy LRU. In this situation, the Longest Unused turns out to be superior to the Nearest Neighbor.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) offline clustering / two-level memory / MIN page replacement / LRU page replacement / heuristic algorithm / dynamic clustering
Paper # COMP97-44
Date of Issue

Conference Information
Committee COMP
Conference Date 1997/9/26(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 Theoretical Foundations of Computing (COMP)
Language ENG
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Towards Dynamic Clustering
Sub Title (in English)
Keyword(1) offline clustering
Keyword(2) two-level memory
Keyword(3) MIN page replacement
Keyword(4) LRU page replacement
Keyword(5) heuristic algorithm
Keyword(6) dynamic clustering
1st Author's Name Jun Kiniwa
1st Author's Affiliation Department of Management Science, Kobe University of Commerce()
2nd Author's Name Tiko Kameda
2nd Author's Affiliation School of computing Science, Simon Fraser University
Date 1997/9/26
Paper # COMP97-44
Volume (vol) vol.97
Number (no) 290
Page pp.pp.-
#Pages 8
Date of Issue