A generic merge-based dynamic indexing framework for iMeMex
Blum Sandro · Repository for Publications and Research Data (ETH Zurich) · 2008
A Personal Dataspace Management System must be able to handle highly dynamic dataspaces. This requires index structures that are not only capable of efficiently speeding up queries on these dataspaces but that can also be updated on-the-fly. State of the art text-retrieval systems are typically based on inverted file indices that are updated either in-place or with a merge-based approach. In this thesis we present a merge-based dynamic indexing framework for the iMeMex Dataspace Management System. The framework we present is generic, extensible and is based on abstract sub components. We provide several sub component implementations and evaluate the system experimentally. Furthermore we have studied analytically the three best-known merge strategies: No merge, Immediate Merge and Logarithmic Merge. Based on this analysis we propose a cost model for determining the best strategy in a given scenario or as the basis for an adaptive merge strategy. Acknowledgments I would like to express my gratitude to my supervisors Dr. JensPeter Dittrich, Marcos Antonio Vaz Salles, and Lukas Blunschi for their provided help, guidance and continuous support through the course of this work. I would also like to thank them and the whole iMeMex group for the very friendly and helpful atmosphere.