Ray Tracing acceleration through a custom scheduling policy to take advantage of the cache affinity in a Linux-based Special-Purpose Operating System
Loading...
Date
Authors
Camacho-Mora, Alvaro
Journal Title
Journal ISSN
Volume Title
Publisher
Instituto Tecnológico de Costa Rica
Abstract
Esta investigación explora el beneficio de diseñar una política de calendarizacion personalizada
que reduzca el tiempo de de ejecución de cargas computacionalmente intensivas.
Cargas computacionalmente intensivas tales como ray tracing, son sensibles al
cambio de contexto producido por el calendarizador. La política de calendarización propuesta
asigna afinidad de cache fuerte para reducir el cambio de contexto al permitir que
cada hilo tenga asignado un único núcleo para su ejecución. Utilizando un sistema operativo
de propósito especifico, hipotéticamente, el sistema tendrá un mayor rendimiento al
combinarlo con la política de calendarización personalizada. El algoritmo de ray tracing
fue seleccionado como carga computacionalmente intensiva para comparar su rendimiento
en un sistema operativo de propósito especifico contra un sistema operativo de propósito
general con su configuración por defecto. Comparado a la referencia, ANOVA factorial
confirmo un 19% de reducción en el tiempo de sintetizado promedio al usar la política de
calendarización personalizada en un sistema operativo de propósito especifico.
The present research explores the benefit of designing a custom scheduling policy to reduce the execution time for computationally intensive workloads. Computationally intensive workloads, such as, ray tracing, are sensible to the context switching produced by the scheduler. The proposed custom scheduling policy assigns hard cache affinity to reduce the context switching by allowing each thread to use only one core during the process execution. Utilizing a special-purpose operating system will hypothetically boost the reduced execution time by integrating the custom scheduling policy. Ray tracing algorithm was selected as the computationally intensive workload to compare its performance in the special-purpose operating system with the custom scheduling policy against a generalpurpose operating system with the default configuration. Compared to the baseline, the factorial ANOVA test confirmed an average 19% reduction of the rendering time using the custom scheduling policy in a special-purpose operating system.
The present research explores the benefit of designing a custom scheduling policy to reduce the execution time for computationally intensive workloads. Computationally intensive workloads, such as, ray tracing, are sensible to the context switching produced by the scheduler. The proposed custom scheduling policy assigns hard cache affinity to reduce the context switching by allowing each thread to use only one core during the process execution. Utilizing a special-purpose operating system will hypothetically boost the reduced execution time by integrating the custom scheduling policy. Ray tracing algorithm was selected as the computationally intensive workload to compare its performance in the special-purpose operating system with the custom scheduling policy against a generalpurpose operating system with the default configuration. Compared to the baseline, the factorial ANOVA test confirmed an average 19% reduction of the rendering time using the custom scheduling policy in a special-purpose operating system.
Description
Proyecto de Graduación (Maestría en Electrónica) Instituto Tecnológico de Costa Rica, Escuela de Ingeniería Electrónica, 2021