Skip to Main content Skip to Navigation

Une nouvelle architecture d'automatisation des réseaux : de la détection d'anomalie à la reconfiguration dynamique

Abstract : Legacy and novel network services are expected to be migrated and designedto be deployed in fully virtualized environments which lead to novel network architecturesthat challenge legacy fault management systems and resilience characterization. Indeed, thecoordination among the different software components for, e.g., orchestration, switching, andvirtual machine and container management creates different monitoring points, besides novelsources of faults and bugs. In this thesis, we propose a network automation framework thatdetects anomalies and characterizes the resiliency state of a virtualized network service. ALong-Short-Term-Memory-Autoencoder-based algorithm analyzes a multidimensional timeseriesbuilt from hundreds of metrics collected at the physical, virtual, and service layers.It learns the nominal working conditions of both the infrastructure and the service, and foreach type of resource (i.e., CPU, network, memory, and disk); it then detects and analyzesdeviations (anomalies) from the learned reference. The produced deviations characterizationis finally used to generate both the transition state graph and the innovative radiographyvisualization. The latter compactly visualizes the propagation of anomalies across all thelayers down from the physical and up to the service, highlighting the temporal evolutionas well. The former aims at establishing the virtualized platform state as the basis for are-orchestration algorithm that leverages a novel reputation-based resiliency managementtechnique. We implement and validate the proposed framework through experimental tests onthe Kubernetes platform hosting a containerized, open-source, and virtualized network core service.
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Friday, July 1, 2022 - 1:06:08 AM
Last modification on : Friday, August 5, 2022 - 2:54:00 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03710828, version 1



Alessio Diamanti. Une nouvelle architecture d'automatisation des réseaux : de la détection d'anomalie à la reconfiguration dynamique. Performance et fiabilité [cs.PF]. HESAM Université, 2021. Français. ⟨NNT : 2021HESAC036⟩. ⟨tel-03710828⟩



Record views


Files downloads