Explotación de técnicas de gestión de recursos en Linux para garantizar aislamiento entre aplicaciones y calidad de servicio en la nube
Loading...
Official URL
Full text at PDC
Publication date
2025
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
En los sistemas multicore actuales, la contención por recursos compartidos como la caché, la memoria, el disco y la red puede degradar significativamente la calidad de servicio ofrecida a las aplicaciones, especialmente en entornos cloud. Aunque los procesadores modernos incluyen soporte hardware para el particionado de caché, permitiendo al sistema operativo (SO) gestionar de forma más justa estos recursos, otros elementos críticos como el disco y la red requieren soluciones adicionales. Este Trabajo de Fin de Grado trata de abordar ambos problemas: por un lado, se propone una técnica de particionado de caché a nivel de kernel para mejorar la equidad y la calidad de servicio en la nube entre aplicaciones concurrentes; por otro, se realiza un estudio y prueba de concepto de herramientas y tecnologías que permiten gestionar la contención en otros recursos, tales como disco y red. Estas últimas operan a nivel
de espacio de usuario, permitiendo una completa independencia y coexistencia con el particionado de caché. Los experimentos realizados utilizando cargas representativas de entornos cloud muestran que la integración de estas estrategias contribuye a mejorar la equidad y la calidad de servicio, siendo especialmente efectiva en escenarios con cargas de trabajo heterogéneas. Mediante monitorización de disco se revela la existencia de problemas de contención. Estos problemas son mitigados gracias a la limitación del uso de disco, permitiendo el mantenimiento de la calidad de servicio.
In current multicore systems, contention for shared resources such as cache, memory, disk, and network can significantly degrade the quality of service offered to applications, especially in cloud environments. Although modern processors include hardware support for cache partitioning, allowing the operating system (OS) to manage these resources more fairly, other critical elements such as disk and network require additional solutions. This TFG aims to address both problems: on the one hand, a kernel-level cache partitioning technique is proposed to improve fairness and quality of service in the cloud among concurrent applications; on the other hand, a study and proof of concept of tools and technologies are carried out to manage contention in other resources, such as disk and network. The latter operate at the user-space level, allowing complete independence from and coexistence with cache partitioning. Experiments conducted using representative cloud environment workloads show that the integration of these strategies contributes to improving fairness and quality of service, being especially effective in scenarios with heterogeneous workloads. Through disk monitoring, the existence of contention problems is revealed. These problems are mitigated through disk usage limitation, allowing the maintenance of quality of service.
In current multicore systems, contention for shared resources such as cache, memory, disk, and network can significantly degrade the quality of service offered to applications, especially in cloud environments. Although modern processors include hardware support for cache partitioning, allowing the operating system (OS) to manage these resources more fairly, other critical elements such as disk and network require additional solutions. This TFG aims to address both problems: on the one hand, a kernel-level cache partitioning technique is proposed to improve fairness and quality of service in the cloud among concurrent applications; on the other hand, a study and proof of concept of tools and technologies are carried out to manage contention in other resources, such as disk and network. The latter operate at the user-space level, allowing complete independence from and coexistence with cache partitioning. Experiments conducted using representative cloud environment workloads show that the integration of these strategies contributes to improving fairness and quality of service, being especially effective in scenarios with heterogeneous workloads. Through disk monitoring, the existence of contention problems is revealed. These problems are mitigated through disk usage limitation, allowing the maintenance of quality of service.
Description
Trabajo de Fin de Grado en Ingeniería de Computadores e Ingeniería Informática, Facultad de Informática UCM, Departamento de Arquitectura de Computadores y Automática, Curso 2024/2025.