View in EDS

Experiences in data-parallel simulation and analysis of complex systems with irregular graph structures : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Computer Science at Massey University, Albany, New Zealand

Saved in:
Bibliographic Details
Title: Experiences in data-parallel simulation and analysis of complex systems with irregular graph structures : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Computer Science at Massey University, Albany, New Zealand
Authors: Leist, Arno
Publisher Information: Massey University
Publication Year: 2011
Collection: Massey University: Massey Research Online
Subject Terms: Parallel processing (Computer science), Graphics processing units, Computer simulation
Description: The interactions between the components of many natural and artificial systems can be described using a graph. These graphs often have an irregular structure with non-trivial topological features. Complex system behaviour emerges on the macroscopic scale from a large number of relatively simple interactions on the microscopic scale. To better understand the observed behaviour of a complex system, the interactions among its basic elements are commonly described in a computational model. As long as the interactions are defined accurately and the number of elements is large enough for complex patterns to emerge, a simulation based on such a model is expected to produce the same behaviour as the system under investigation. The difficulty is often to simulate the model on a large enough scale to obtain scientifically meaningful results. Powerful computer systems are required to calculate the effects caused by the interactions of large numbers of elements. Supercomputers that are constructed from hundreds of thousands of processing units can be used to update many components of the system in parallel and thus reduce the overall simulation time, but these systems are expensive to buy and maintain. As the processor architectures used in workstations and regular desktop computers are becoming more powerful, a small cluster constructed from these systems can be a more viable option. In recent years, the highly data parallel architecture of commodity graphics processing units (GPUs) has received a growing amount of attention due to their high peak compute throughput compared to central processing units (CPUs). New software development tools that turn the GPU hardware into a general purpose compute accelerator have become available. This thesis describes how GPUs can be used to accelerate scientific simulations of complex systems that are based on irregular graph structures. New software development approaches and algorithms are needed to fully utilise the data parallel many-core architecture of today’s GPUs. Irregular ...
Document Type: thesis
Language: English
Relation: http://hdl.handle.net/10179/2992; Q112886912
Availability: http://hdl.handle.net/10179/2992
Rights: The Author
Accession Number: edsbas.C59A5FAA
Database: BASE
Description
Abstract:The interactions between the components of many natural and artificial systems can be described using a graph. These graphs often have an irregular structure with non-trivial topological features. Complex system behaviour emerges on the macroscopic scale from a large number of relatively simple interactions on the microscopic scale. To better understand the observed behaviour of a complex system, the interactions among its basic elements are commonly described in a computational model. As long as the interactions are defined accurately and the number of elements is large enough for complex patterns to emerge, a simulation based on such a model is expected to produce the same behaviour as the system under investigation. The difficulty is often to simulate the model on a large enough scale to obtain scientifically meaningful results. Powerful computer systems are required to calculate the effects caused by the interactions of large numbers of elements. Supercomputers that are constructed from hundreds of thousands of processing units can be used to update many components of the system in parallel and thus reduce the overall simulation time, but these systems are expensive to buy and maintain. As the processor architectures used in workstations and regular desktop computers are becoming more powerful, a small cluster constructed from these systems can be a more viable option. In recent years, the highly data parallel architecture of commodity graphics processing units (GPUs) has received a growing amount of attention due to their high peak compute throughput compared to central processing units (CPUs). New software development tools that turn the GPU hardware into a general purpose compute accelerator have become available. This thesis describes how GPUs can be used to accelerate scientific simulations of complex systems that are based on irregular graph structures. New software development approaches and algorithms are needed to fully utilise the data parallel many-core architecture of today’s GPUs. Irregular ...