A mechanism to detect changing access patterns and automatically migrate distributed R-tree indexed multidimensional data

Abstract

We present an algorithm to migrate multidimensional data in a non-replicated distributed environment. Our proposed algorithm is intented to improve query performance for mobile objects. Our algorithm automatically detects access pattern changes and migrates portions of the data from current sites of residence to sites recently accessing the data most frequently, thus reducing remote communication costs. We assume the data is indexed by an R-tree multidimensional index and that a global R-tree is used to located and retrieve portions of the data set. We present a distributed R-tree structure and experimentally explore a specific access detection and migration mechanism. Our experimental results show that when compared to the no-migration case, and depending on network speed, our migration scheme may result in a query time reduction of a few percent to an order of magnitude.