Squirrel: Architecture Driven Resource Management - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2016

Squirrel: Architecture Driven Resource Management

Résumé

Resource management is critical to guarantee Quality of Service when various stakeholders share the execution environment , such as cloud or mobile environments. In this context, providing management techniques compatible with standard practices, such as component models, is essential. Resource management is often realized through monitoring or process isolation (using virtual machines or system containers). These techniques (i) impose varying levels of overhead depending on the managed resource, and (ii) are applied at different abstraction levels, such as processes, threads or objects. Thus, mapping components to system-level abstractions in the presence of resource management requirements can lead to sub-optimal systems. We propose Squirrel, an approach to tune component deployment and resource management in order to reduce management overhead. At run-time, Squirrel uses an architectural model annotated with resource requirements to guide the mapping of components to system abstractions, providing different resource management capabilities and overhead. We present an implementation of Squirrel, using a Java component framework, and a set of experiments to validate its feasibility and overhead. We show that choosing the right component-to-system mappings at deployment-time reduces performance penalty and/or volatile main memory use.
Fichier principal
Vignette du fichier
typeinst.pdf (416.13 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-01355000 , version 1 (22-08-2016)

Identifiants

Citer

Inti Gonzalez-Herrera, Johann Bourcier, Walter Rudametkin, Olivier Barais, Francois Fouquet. Squirrel: Architecture Driven Resource Management. SAC - 31st Annual ACM Symposium on Applied Computing, Apr 2016, Pisa, Italy. ⟨10.1145/0000000.0000000⟩. ⟨hal-01355000⟩
812 Consultations
152 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More