A parallel source finder framework and Gaussian filter implementation for large radio astronomy spectroscopic data using high performance computing

Stefan Westerlund

    Research output: ThesisDoctoral Thesis

    404 Downloads (Pure)

    Abstract

    [Truncated] New radio astronomy telescopes, such as the Australian Square Kilometre Array Path nder (ASKAP), have a larger eld of view and greater resolution than previous instruments. This extra information signifficantly increases the amount of image data produced, up to petabytes in size for an all-sky survey. Surveys that use these telescopes will need to search these large images for galaxies and other sources of emission. Existing source finders, which run on consumer desktop or laptop computers, are unable to process this volume of data at the same speed at which the images are produced. New source finding programs are needed that can use additional processing hardware to perform searches in a reasonable amount of time. It is also desirable to reduce the computational requirements of source finding, such that the processing time saved can be used for more intensive algorithms elsewhere in the survey processing, or to reduce the computational costs of the survey.
    This thesis examines the development of a framework for parallel source finders for neutral hydrogen (HI) data, and a parallel source finding program that is capable of searching large spectroscopic images using High Performance Computing (HPC) techniques. Modern HPC hardware consists of a large number of processors connected by a network, and for a program to make use of the processing power available it must be written so that it is able to execute on each processing core. This necessitates writing the program to work in parallel, and efficiently dividing work between the different processors. However, it is non-trivial to create a parallel version of a single-threaded algorithm. In this work I have produced the Source Finder Accuracy Evaluator (SFAE), a program that measures the accuracy of source finders; the Scalable Source Finding Framework (SSoFF), created to ease the development of parallel source finders; and the Parallel Gaussian Source Finder (PGSF), a parallel source finding program developed using SSoFF.
    Original languageEnglish
    QualificationDoctor of Philosophy
    Publication statusUnpublished - 2014

    Fingerprint

    Radio astronomy
    Processing
    Telescopes
    Hardware
    Laptop computers
    Galaxies
    Personal computers
    Hydrogen
    Costs

    Cite this

    @phdthesis{431f7f8088224236b297034084d95557,
    title = "A parallel source finder framework and Gaussian filter implementation for large radio astronomy spectroscopic data using high performance computing",
    abstract = "[Truncated] New radio astronomy telescopes, such as the Australian Square Kilometre Array Path nder (ASKAP), have a larger eld of view and greater resolution than previous instruments. This extra information signifficantly increases the amount of image data produced, up to petabytes in size for an all-sky survey. Surveys that use these telescopes will need to search these large images for galaxies and other sources of emission. Existing source finders, which run on consumer desktop or laptop computers, are unable to process this volume of data at the same speed at which the images are produced. New source finding programs are needed that can use additional processing hardware to perform searches in a reasonable amount of time. It is also desirable to reduce the computational requirements of source finding, such that the processing time saved can be used for more intensive algorithms elsewhere in the survey processing, or to reduce the computational costs of the survey. This thesis examines the development of a framework for parallel source finders for neutral hydrogen (HI) data, and a parallel source finding program that is capable of searching large spectroscopic images using High Performance Computing (HPC) techniques. Modern HPC hardware consists of a large number of processors connected by a network, and for a program to make use of the processing power available it must be written so that it is able to execute on each processing core. This necessitates writing the program to work in parallel, and efficiently dividing work between the different processors. However, it is non-trivial to create a parallel version of a single-threaded algorithm. In this work I have produced the Source Finder Accuracy Evaluator (SFAE), a program that measures the accuracy of source finders; the Scalable Source Finding Framework (SSoFF), created to ease the development of parallel source finders; and the Parallel Gaussian Source Finder (PGSF), a parallel source finding program developed using SSoFF.",
    keywords = "Radio astronomy, Source finding, High performance computing, Parallel computing, GPU computing, Image processing",
    author = "Stefan Westerlund",
    year = "2014",
    language = "English",

    }

    TY - THES

    T1 - A parallel source finder framework and Gaussian filter implementation for large radio astronomy spectroscopic data using high performance computing

    AU - Westerlund, Stefan

    PY - 2014

    Y1 - 2014

    N2 - [Truncated] New radio astronomy telescopes, such as the Australian Square Kilometre Array Path nder (ASKAP), have a larger eld of view and greater resolution than previous instruments. This extra information signifficantly increases the amount of image data produced, up to petabytes in size for an all-sky survey. Surveys that use these telescopes will need to search these large images for galaxies and other sources of emission. Existing source finders, which run on consumer desktop or laptop computers, are unable to process this volume of data at the same speed at which the images are produced. New source finding programs are needed that can use additional processing hardware to perform searches in a reasonable amount of time. It is also desirable to reduce the computational requirements of source finding, such that the processing time saved can be used for more intensive algorithms elsewhere in the survey processing, or to reduce the computational costs of the survey. This thesis examines the development of a framework for parallel source finders for neutral hydrogen (HI) data, and a parallel source finding program that is capable of searching large spectroscopic images using High Performance Computing (HPC) techniques. Modern HPC hardware consists of a large number of processors connected by a network, and for a program to make use of the processing power available it must be written so that it is able to execute on each processing core. This necessitates writing the program to work in parallel, and efficiently dividing work between the different processors. However, it is non-trivial to create a parallel version of a single-threaded algorithm. In this work I have produced the Source Finder Accuracy Evaluator (SFAE), a program that measures the accuracy of source finders; the Scalable Source Finding Framework (SSoFF), created to ease the development of parallel source finders; and the Parallel Gaussian Source Finder (PGSF), a parallel source finding program developed using SSoFF.

    AB - [Truncated] New radio astronomy telescopes, such as the Australian Square Kilometre Array Path nder (ASKAP), have a larger eld of view and greater resolution than previous instruments. This extra information signifficantly increases the amount of image data produced, up to petabytes in size for an all-sky survey. Surveys that use these telescopes will need to search these large images for galaxies and other sources of emission. Existing source finders, which run on consumer desktop or laptop computers, are unable to process this volume of data at the same speed at which the images are produced. New source finding programs are needed that can use additional processing hardware to perform searches in a reasonable amount of time. It is also desirable to reduce the computational requirements of source finding, such that the processing time saved can be used for more intensive algorithms elsewhere in the survey processing, or to reduce the computational costs of the survey. This thesis examines the development of a framework for parallel source finders for neutral hydrogen (HI) data, and a parallel source finding program that is capable of searching large spectroscopic images using High Performance Computing (HPC) techniques. Modern HPC hardware consists of a large number of processors connected by a network, and for a program to make use of the processing power available it must be written so that it is able to execute on each processing core. This necessitates writing the program to work in parallel, and efficiently dividing work between the different processors. However, it is non-trivial to create a parallel version of a single-threaded algorithm. In this work I have produced the Source Finder Accuracy Evaluator (SFAE), a program that measures the accuracy of source finders; the Scalable Source Finding Framework (SSoFF), created to ease the development of parallel source finders; and the Parallel Gaussian Source Finder (PGSF), a parallel source finding program developed using SSoFF.

    KW - Radio astronomy

    KW - Source finding

    KW - High performance computing

    KW - Parallel computing

    KW - GPU computing

    KW - Image processing

    M3 - Doctoral Thesis

    ER -