Resumen: In container management systems, such as Kubernetes, the scheduler has to place containers in physical machines and it should be aware of the degradation in performance caused by placing together containers that are barely isolated. We propose that clients provide a characterization of their applications to allow a scheduler to evaluate what is the best configuration to deal with the workload at a given moment. The default Kubernetes Scheduler only takes into account the sum of requested resources in each machine, which is insufficient to deal with the performance degradation. In this paper, we show how specifying resource limits is not enough to avoid resource contention, and we propose the architecture of a scheduler, based on the client application characterization, to avoid the resource contention. Idioma: Inglés DOI: 10.1007/978-3-319-68066-8_13 Año: 2017 Publicado en: Lecture Notes in Computer Science 10537 (2017), 162-176 ISSN: 0302-9743 Financiación: info:eu-repo/grantAgreement/ES/DGA/T93 Financiación: info:eu-repo/grantAgreement/ES/MINECO/TIN2013-40809-R Tipo y forma: Article (PostPrint) Área (Departamento): Área Arquit.Tecnología Comput. (Dpto. Informát.Ingenie.Sistms.) Área (Departamento): Área Lenguajes y Sistemas Inf. (Dpto. Informát.Ingenie.Sistms.)
Exportado de SIDERAL (2024-02-02-14:57:36)