The Library
Data structure abstraction and parallelisation of multi-material hydrodynamic applications
Tools
Kirk, Richard Oliver (2020) Data structure abstraction and parallelisation of multi-material hydrodynamic applications. PhD thesis, University of Warwick.
|
PDF
WRAP_Theses_Kirk_2020.pdf - Submitted Version - Requires a PDF viewer. Download (1783Kb) | Preview |
Official URL: http://webcat.warwick.ac.uk/record=b371784
Abstract
The aim for High Performance Computing (HPC) is to achieve the best performance for an application, in order to execute it as quickly as possible. This is often achieved through iterative improvements in Central Processing Unit (CPU) technology such as: including more circuitry by shrinking or making processors larger; making the processor run faster by increasing the clock speed; or increasing the amount of parallelism. Recently, there has been increasing diversity in how HPC systems achieve these performance improvements. The use of Graphics Processing Unit (GPU) processors has become more common, and there has been a growing interest in high bandwidth memory. This has lead to a need for performance portable code, so programs may be written once but compiled and ran on a range of differing systems, with minimal impact on the performance.
As memory becomes a major focus, so too should the data structure used by an application. Without a well designed data structure, the performance of a program can be affected. However, it is key that this is done in a performance portable way, where the data structure can be altered and optimised without the need for the application to be rewritten. As such, a data structure abstraction library was developed, calledWarwick Data Store (WDS). This library is able to provide objects, which allow for access to data, without the application needing to know the detail of the data structure. The library also provides additional functionality that would otherwise be difficult and time consuming to implement, such as the ability to convert a variable or a collection of variables from one data structure to another. The performance impact of the library is shown to be minimal, especially in larger problem sizes. Because of the flexibility of the library, data structures for specialised cases can be implemented into WDS without impacting the performance of other data structures. The performance of these specialised data structures is also presented as being minimal.
Item Type: | Thesis (PhD) | ||||
---|---|---|---|---|---|
Subjects: | Q Science > QA Mathematics > QA76 Electronic computers. Computer science. Computer software | ||||
Library of Congress Subject Headings (LCSH): | High performance computing, Data structures (Computer science), Electronic data processing -- Structured techniques, Abstract data types (Computer science), Computer architecture | ||||
Official Date: | September 2020 | ||||
Dates: |
|
||||
Institution: | University of Warwick | ||||
Theses Department: | Department of Computer Science | ||||
Thesis Type: | PhD | ||||
Publication Status: | Unpublished | ||||
Copyright Holders: | Crown copyright | ||||
Supervisor(s)/Advisor: | Jarvis, Stephen A., 1970- ; Mudalige, Gihan R. | ||||
Sponsors: | UK Atomic Weapons Establishment (AWE plc) | ||||
Format of File: | |||||
Extent: | xxi, 168 leaves : illustrations | ||||
Language: | Eng |
Request changes or add full text files to a record
Repository staff actions (login required)
View Item |
Downloads
Downloads per month over past year