Dickson, James (2018) Towards application-centric I/O benchmarking for parallel scientific applications. PhD thesis, University of Warwick.

Preview

PDF WRAP_Theses_Dickson_2018.pdf - Submitted Version - Requires a PDF viewer. Download (4Mb) | Preview

Request Changes to record.

Abstract

High performance computing (HPC) systems are undergoing an explosion in the variation and complexity of their hardware components and architectures. By the early 2020s, it is predicted that exascale systems will be in operation, and it is the pursuit of this capability that is currently shifting the parallel computing landscape.

Storage systems have developed alongside the other supercomputing components, but have struggled to keep pace with the rate of progress of computational hardware in particular. The resulting performance bottlenecks are problematic, as the data storage demands for the purposes of analysis and resilience are unlikely to be reduced by the current and future generations of HPC platforms.

The principle focus of the work presented in this thesis is to enable the benchmarking and analysis of I/O in a meaningful way, such that the performance expected from a scientific application can be accurately measured. Specifically, this thesis presents a case study of the profiling of multi-physics workloads for the purpose of extracting workload characteristics that are not limited by commercial sensitivity, as with the application itself. A flexible and portable I/O proxy application is developed and validated, before being deployed with the previously defined workloads to benchmark I/O performance on a number of current generation systems. This thesis ends with an analysis of the replicated workloads’ sensitivity to the configuration of different elements of the I/O software stack. An evaluation is also performed on some alternative high level I/O implementation strategies that can be adopted to simplify the adoption of new burst buffer storage architectures.

Item Type:	Thesis (PhD)
Subjects:	Q Science > QA Mathematics > QA76 Electronic computers. Computer science. Computer software
Library of Congress Subject Headings (LCSH):	Computer input-output equipment, Computer storage devices, High performance computing
Official Date:	September 2018
Dates:	Date Event September 2018 UNSPECIFIED
Institution:	University of Warwick
Theses Department:	Department of Computer Science
Thesis Type:	PhD
Publication Status:	Unpublished
Supervisor(s)/Advisor:	Jarvis, Stephen A., 1970-
Sponsors:	Atomic Weapons Establishment (Great Britain) ; Engineering and Physical Sciences Research Council
Format of File:	pdf
Extent:	xxi, 171 leaves : illustrations
Language:	eng

Request changes or add full text files to a record

Repository staff actions (login required)

View Item

Downloads