Improving heterogenous storage performance in HPC Cloud systems using efficient storage algorithms informed by statistical models

Scientific applications are widely used to solve complex problems from different do- mains. These kinds of applications usually have demanding computational require- ments. Hence they must be executed in HPC clusters to guarantee a successful execution and find an optimal solution. In the last years, researchers have tried to find an alternative to run their applications in cloud computing. Recent works have been attempting to migrate the applications because they see a flexibility and sca- lability model in cloud computing that can benefit them and their applications. The cloud computing economic model, where you only pay for what you are using, can reduce the cost of the acquisition, maintenance, and updates in comparison with a HPC cluster. The deployment of HPC applications over cloud computing clusters presents several challenges that have yet to be resolved. One potential problem con- cerns storage systems and file systems, as cloud clusters do not use the same sto- rage and file systems as HPC clusters. Therefore, HPC applications are affected by overheads given by the different technologies and the entire environment. This dis- sertation seeks to reduce HPC applications’ overhead, improving the performance of applications running on heterogeneous storage systems in the HPC Cloud. To do so, this dissertation characterizes the performance of High Performance Computing ap- plications that make use of heterogeneous storage technologies in cloud computing clusters. This dissertation also presents and validates the use of an Extreme Value Theory-based model to characterize, analyze and predict the performance of these applications. Finally, this dissertation presents a genetic algorithm that uses the pro- posed model as input to solve an Integer Linear Programming problem formulated for the data placement of the files used by the applications to the heterogeneous storage devices in a HPC cloud system.

• Autores: Marquez Franco, Jack Daniels
• Tipo de recurso: Doctoral thesis
• Fecha de publicación: 2022
• Institución: Universidad Autónoma de Occidente
• Repositorio: RED: Repositorio Educativo Digital UAO
• Idioma: eng
• OAI Identifier: oai:red.uao.edu.co:10614/13693
• Acceso en línea: https://hdl.handle.net/10614/13693; https://red.uao.edu.co/
• Palabra clave: Doctorado en Ingeniería; Computación en la nube; Algoritmos genéticos; Computación evolutiva; Cloud computing; Evolutionary computation; HPC Cloud; EVT; Genetic algorithm; Heterogeneous storage
• Rights: openAccess
• License: Derechos reservados - Universidad Autónoma de Occidente, 2022