Experimental Studies of Heuristic Algorithms for Two Level Object Clustering

Abstract

The clustering of objects in a layered object storage system is by common consent an exceedingly difficult problem. Studies the performance of three heuristic placement algorithms. A series of eight reasonably realistic case studies were used as a benchmark battery, and several hundred experiments were carried out to evaluate results of using the algorithms. Presents the results and the insights gained from the study.

An experiment in knowledge discovery using data dependencies

“…In our physical database design study, in order to improve the performance of relational database systems (Bell, 1984), clustering was used to optimize the physical proximity of interrelated tuples on secondary storage devices. Simulated annealing (Bell et al, 1990;Kirkpatrick et al, 1983), graph collapsing (Bell et al, 1988), and genetic adaptation (Bell and McErlean, 1994) were compared in our previous studies. Most clustering algorithms use a measure of similarity to create clusters of associated terms.…”

Classification of knee problems to provide evidential support

“…In our physical database design study in order to improve the performance of relational database systems (Bell, 1984), clustering was used to optimize the physical proximity of interrelated tuples on secondary storage devices. Simulated annealing (Kirkpatrick et al, 1983;Bell et al, 1990), graph collapsing (Bell et al, 1988), and genetic adaptation (Bell and McErlean, 1994) were compared in our previous studies. Most clustering algorithms use a measure of similarity to create clusters of associated terms.…”