Graph database management systems: storage, management and query processing

Goonetilleke, T 2017, Graph database management systems: storage, management and query processing, Doctor of Philosophy (PhD), Science, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Goonetilleke.pdf Thesis application/pdf 3.38MB
Title Graph database management systems: storage, management and query processing
Author(s) Goonetilleke, T
Year 2017
Abstract The proliferation of graph data, generated from diverse sources, have given rise to many research efforts concerning graph analysis. Interactions in social networks, publication networks, protein networks, software code dependencies and transportation systems are all examples of graph-structured data originating from a variety of application domains and demonstrating different characteristics. In recent years, graph database management systems (GDBMS) have been introduced for the management and analysis of graph data. Motivated by the growing number of real-life applications making use of graph database systems, this thesis focuses on the effectiveness and efficiency aspects of such systems. Specifically, we study the following topics relevant to graph database systems: (i) modeling large-scale applications in GDBMS; (ii) storage and indexing issues in GDBMS, and (iii) efficient query processing in GDBMS. In this thesis, we adopt two different application scenarios to examine how graph database systems can model complex features and perform relevant queries on each of them. Motivated by the popular application of social network analytics, we selected Twitter, a microblogging platform, to conduct our detailed analysis. Addressing limitations of existing models, we pro- pose a data model for the Twittersphere that proactively captures Twitter-specific interactions. We examine the feasibility of running analytical queries on GDBMS and offer empirical analysis of the performance of the proposed approach. Next, we consider a use case of modeling software code dependencies in a graph database system, and investigate how these systems can support capturing the evolution of a codebase overtime. We study a code comprehension tool that extracts software dependencies and stores them in a graph database. On a versioned graph built using a very large codebase, we demonstrate how existing code comprehension queries can be efficiently processed and also show the benefit of running queries across multiple versions. Another important aspect of this thesis is the study of storage aspects of graph systems. Throughput of many graph queries can be significantly affected by disk I/O performance; therefore graph database systems need to focus on effective graph storage for optimising disk operations. We observe that the locality of edges plays an important role and we address the edge-labeling problem which aims to label both incoming and outgoing edges of a graph maximizing the ‘edge-consecutiveness’ metric. By achieving a better layout and locality of edges on disk, we show that our proposed algorithms result in significantly improved disk I/O performance leading to faster execution of neighbourhood queries. Some applications require the integrated processing of queries from graph and the textual domains within a graph database system. Aggregation of these dimensions facilitates gaining key insights in several application scenarios. For example, in a social network setting, one may want to find the closest k users in the network (graph traversal) who talk about a particular topic A (textual search). Motivated by such practical use cases, in this thesis we study the top-k social-textual ranking query that essentially requires efficient combination of a keyword search query with a graph traversal. We propose algorithms that leverage graph partitioning techniques, based on the premise that socially close users will be placed within the same partition, allowing more localised computations. We show that our proposed approaches are able to achieve significantly better results compared to standard baselines and demonstrating robust behaviour under changing parameters.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Science
Subjects Information Systems Management
Information Retrieval and Web Search
Database Management
Keyword(s) Graph database management systems
Graph storage
Graph data model
Graph query processing
Version Filter Type
Access Statistics: 179 Abstract Views, 400 File Downloads  -  Detailed Statistics
Created: Wed, 04 Apr 2018, 16:07:58 EST by Adam Rivett
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us