Cheng, Long (2014) A scalable analysis framework for large-scale RDF data. PhD thesis, National University of Ireland Maynooth.

Preview

Download (2MB) | Preview

Abstract

With the growth of the Semantic Web, the availability of RDF datasets from multiple domains as Linked Data has taken the corpora of this web to a terabyte-scale, and challenges modern knowledge storage and discovery techniques. Research and engineering on RDF data management systems is a very active area with many standalone systems being introduced. However, as the size of RDF data increases, such single-machine approaches meet performance bottlenecks, in terms of both data loading and querying, due to the limited parallelism inherent to symmetric multi-threaded systems and the limited available system I/O and system memory. Although several approaches for distributed RDF data processing have been proposed, along with clustered versions of more traditional approaches, their techniques are limited by the trade-off they exploit between loading complexity and query efficiency in the presence of big RDF data. This thesis then, introduces a scalable analysis framework for processing large-scale RDF data, which focuses on various techniques to reduce inter-machine communication, computation and load-imbalancing so as to achieve fast data loading and querying on distributed infrastructures. The first part of this thesis focuses on the study of RDF store implementation and parallel hashing on big data processing. (1) A system-level investigation of RDF store implementation has been conducted on the basis of a comparative analysis of runtime characteristics of a representative set of RDF stores. The detailed time cost and system consumption is measured for data loading and querying so as to provide insight into different triple store implementation as well as an understanding of performance differences between different platforms. (2) A high-level structured parallel hashing approach over distributed memory is proposed and theoretically analyzed. The detailed performance of hashing implementations using different lock-free strategies has been characterized through extensive experiments, thereby allowing system developers to make a more informed choice for the implementation of their high-performance analytical data processing systems. The second part of this thesis proposes three main techniques for fast processing of large RDF data within the proposed framework. (1) A very efficient parallel dictionary encoding algorithm, to avoid unnecessary disk-space consumption and reduce computational complexity of query execution. The presented implementation has achieved notable speedups compared to the state-of-art method and also has achieved excellent scalability. (2) Several novel parallel join algorithms, to efficiently handle skew over large data during query processing. The approaches have achieved good load balancing and have been demonstrated to be faster than the state-of-art techniques in both theoretical and experimental comparisons. (3) A two-tier dynamic indexing approach for processing SPARQL queries has been devised which keeps loading times low and decreases or in some instances removes intermachine data movement for subsequent queries that contain the same graph patterns. The results demonstrate that this design can load data at least an order of magnitude faster than a clustered store operating in RAM while remaining within an interactive range for query processing and even outperforms current systems for various queries.

Item Type:	Thesis (PhD)
Keywords:	large-scale RDF data;
Academic Unit:	Faculty of Science and Engineering > Electronic Engineering
Item ID:	5442
Depositing User:	IR eTheses
Date Deposited:	30 Sep 2014 09:56
URI:
Use Licence:	This item is available under a Creative Commons Attribution Non Commercial Share Alike Licence (CC BY-NC-SA). Details of this licence are available here

Repository Staff Only(login required)

Item control page

Downloads

Origin of downloads