Consumo energético de entornos serverless basados en contenedores
Loading...
Official URL
Full text at PDC
Publication date
2023
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
La configuración ideal de cualquier entorno de producción para reducir la carga energética y costes es un tema recurrente en cualquier empresa. Es necesario que todas las tecnologías que componen los sistemas de una empresa tengan la mejor configuración posible o disponible.
Las arquitecturas sin servidor o serverless, son algunas de las nuevas tecnologías recientemente ofrecidas como una alternativa barata de pago por uso cuyo enfoque principal es reducir costes monetarios. Sin embargo, a menudo se deja en segundo plano su coste energético.
En este trabajo, se realiza una aproximación a un entorno real, investigando los elementos que componen un entorno típico sin servidor y analizando sus particularidades de consumo.
Se escala en un primer nivel comenzando por el consumo de un contenedor. Se utilizan herramientas como HwPC-Sensor, SmartWatts y PowerAPI, todas basadas en la infraestructura de Intel RAPL, una interfaz que permite acceder a los eventos de consumo de potencia del procesador. El segundo nivel está enfocado en realizar un estudio del impacto de estos contenedores en un entorno físico. Se ejecutan funciones sobre ellos, con librerías de estrés como Stress-NG. Por último, se programa un simulador de un entorno serverless que permite extrapolar conclusiones a entornos reales, variando sus parámetros y obteniendo conclusiones sobre las con guraciones más rentables desde el punto de vista del consumo.
The ideal configuration of any production environment to reduce the energy load and costs is a recurring theme in any company. It is necessary that all the technologies that make up a company's systems have the best possible or available configuration. Serverless architectures are some of the new technologies recently offered as an inexpensive pay-as-you-go alternative whose main focus is to reduce monetary costs. However, their energy cost is often neglected. In this work, a real environment approach is performed, investigating the elements that compose a typical serverless environment and analyzing its consumption particularities. It is scaled at a first level starting with the consumption of a container. Tools such as HwPC-Sensor, SmartWatts and PowerAPI are used, all based on the Intel RAPL infrastructure, an interface that allows access to processor power consumption events. The second level is focused on performing a study of the impact of these containers in a physical environment. Functions are executed on them, with stress libraries such as Stress-NG. Finally, a simulator of a serverless environment is programmed to extrapolate conclusions to real environments, varying its parameters and obtaining conclusions on the most profitable configurations of consumption.
The ideal configuration of any production environment to reduce the energy load and costs is a recurring theme in any company. It is necessary that all the technologies that make up a company's systems have the best possible or available configuration. Serverless architectures are some of the new technologies recently offered as an inexpensive pay-as-you-go alternative whose main focus is to reduce monetary costs. However, their energy cost is often neglected. In this work, a real environment approach is performed, investigating the elements that compose a typical serverless environment and analyzing its consumption particularities. It is scaled at a first level starting with the consumption of a container. Tools such as HwPC-Sensor, SmartWatts and PowerAPI are used, all based on the Intel RAPL infrastructure, an interface that allows access to processor power consumption events. The second level is focused on performing a study of the impact of these containers in a physical environment. Functions are executed on them, with stress libraries such as Stress-NG. Finally, a simulator of a serverless environment is programmed to extrapolate conclusions to real environments, varying its parameters and obtaining conclusions on the most profitable configurations of consumption.
Description
Trabajo de Fin de Grado en Ingeniería del Software, Facultad de Informática UCM, Departamento de de Arquitectura de Computadores y Automática, Curso 2022/20233.