Implementing information retrieval using a combined object-oriented database/file system paradigm.

摘要

The SHORE (Scalable Heterogeneous Object REpository) persistent object store under development at the University of Wisconsin-Madison is a hybrid of Object-Oriented Database Management System (OODBMS) and file system technologies. A primary goal of the SHORE project is to provide a system that facilitates the construction of high functionality data servers for a wide variety of applications. In this thesis, I evaluate the suitability of the SHORE system for supporting the implementation of an information retrieval server and evaluate four different distributions of function between client and server. To evaluate the tradeoffs in word index design, two architectures explore an object index versus a server index service. To evaluate client versus server placement of information retrieval functions, two other architectures explore a generic SHORE server versus a custom information retrieval server. My experience with an implementation of a prototype multimedia information server, Chrysalis, demonstrates that overall the SHORE persistent object store functionality is well suited for supporting the construction of an information retrieval server compared with the traditional alternatives of file systems, relational database systems, or even current OODBMS. Several performance studies evaluate major architectural and implementation tradeoffs. Areas for improvement for SHORE are highlighted. Chrysalis is the first major SHORE application, so this dissertation also serves as an experience guide for other potential SHORE applications.