Efficient Algorithms for Querying Large-Scale Data in Relational, XML, and Graph-Structured Data Repositories

Abstract

We live in an information age, and data are ubiquitous today. Various applications, ranging from scientific computing, medical research, and bioinformatics to administrative management, commercial sales, and financial marketing, generate and utilize data every day. Many of these applications are data intensive, with the amount of data involved potentially reaching hundreds of thousands of gigabytes. Further, different applications store data using different data models. For example, applications could store and manage structured data using a flat (relational) model, semi-structured data using a hierarchical (XML) model, and less-structured data using a more general and flexible graph model. In this thesis, I report my research results on efficiently querying large-scale data in relational, XML, and graph-structured data repositories.

Specifically, this thesis covers three research projects, which I have been invited to present in the ACM SIGMOD conference in 2006, 2007, and 2008, respectively. The first project concerns efficient querying of relational data using materialized views and introduces our efficient view-based query-optimization algorithms that support a large and practically important subset of SQL queries. The second project focuses on efficiently querying XML data and presents efficient algorithms for evaluating XPath queries over XML streams, which are the first ones that achieve the O(|D||Q|) time performance, where |D| is the XML data size and |Q| is the XPath query size. Meanwhile, our algorithm EQ also achieves optimal space performance. The third project addresses efficient querying of graph-structured data, by introducing efficient algorithms for retrieving top-ranked tree-pattern matches from large graphs. While a tree-pattern query could have an extremely large, potentially exponential, number of answer matches in a graph, our algorithms exhibit time and space performance that is linear or sub-linear in the size of the input data. Our algorithms are the first ones that have this excellent performance property.