Higher-order particle representation for a portable unstructured particle-in-cell application

[thumbnail of WRAP_Theses_Brown_2020.pdf]
Preview
PDF
WRAP_Theses_Brown_2020.pdf - Submitted Version - Requires a PDF viewer.

Download (2MB) | Preview

Request Changes to record.

Abstract

As the field of High Performance Computing (HPC) moves towards the era of Exascale computation, computer hardware is becoming increasingly parallel and continues to diversify. As a result, it is now crucial for scientific codes to be able to take advantage of a wide variety of hardware types. Additionally, the growth in compute performance has outpaced the improvement in memory latency and bandwidth; this issue now poses a significant obstacle to performance.

This thesis examines these matters in the context of modern plasma physics simulations, specifically those that make use of the Particle-in-Cell (PIC) method on unstructured computational grids. Specifically, we begin by documenting the implementation of the particle-based kernels of such a code using a performance portability library to enable the application to run on a variety of modern hardware, including both CPUs and GPUs. The use of hardware specific tuning is also explored, culminating in a 3x speedup of a key component of the core PIC algorithm. We also show that portability is achievable on both single-node machines and production supercomputers of multiple hardware types.
This thesis also documents an algorithmic change to particle representation within the same code that improves solution accuracy, and adds compute intensity { an important property where memory bandwidth is limited and the ratio of the amount of computation to memory accesses is low. We conclude the work by comparing the performance of the modified algorithm to the base implementation, where we find that shifting the simulation workload towards computation can improve parallel efficiency by up to 2:5x. While the performance improvements that were hoped for were not achieved, we end this thesis by postulating that the proposed methods will become more viable as compilers and hardware improve.

Item Type: Thesis [via Doctoral College] (PhD)
Subjects: Q Science > QA Mathematics > QA76 Electronic computers. Computer science. Computer software
Q Science > QC Physics
Library of Congress Subject Headings (LCSH): High performance computing, Laser plasmas -- Simulation methods, Scattering (Physics), Particle dynamics -- Simulation methods
Official Date: October 2020
Dates:
Date
Event
October 2020
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: Jarvis, Stephen A., 1970- ; Atomic Weapons Establishment (Great Britain) ; Engineering and Physical Sciences Research Council
Format of File: pdf
Extent: xxi, 170 leaves : illustrations, charts
Language: eng
URI: https://wrap.warwick.ac.uk/177400/

Export / Share Citation


Request changes or add full text files to a record

Repository staff actions (login required)

View Item View Item