(Français) Projet SPIREC

Multi-level supervision and prediction for geo-distributed, heterogeneous infrastructures in the Cloud/Edge/IoT continuum

— Preview

Define methods, using distributed machine learning in particular, to enable their efficient management, provide means of securing them and ensure a variety of QoS properties.

Mario Südholt, Professor IMT

Keywords : Supervision, prediction, mitigation; heterogeneous multi-level infrastructures, applications and software stacks; AI, Cloud/Edge/Internet of Things continuum

Today, the cloud is the main infrastructure on which the world’s largest distributed applications are built. New requirements are emerging for computing and storage capacities at the edge, and for the many Internet of Things (IoT) devices that are part of smart cities, the industry of the future, vehicular networks… The resulting cyber ecosystem, the Cloud-Edge-IoT continuum, is characterized by highly heterogeneous infrastructures as well as multi-layered applications and software stacks. Monitoring infrastructures and applications, detecting anomalies in service and application execution, and predicting resource utilization are fundamental services for the continuum. They help to secure applications and ensure numerous quality-of-service properties. However, growing heterogeneity and the use of increasingly complex software stacks require new methods for developing these services. The SPIREC project will address the challenges of monitoring continuum services, detecting their execution anomalies and predicting their resource utilization.

The partners will also develop software components and tools to integrate these functionalities into existing infrastructures and applications, in particular SLICES and future software ecosystems.

— Missions

— Our researches

Data collection and modeling for supervision and anomaly detection

Definition of a hierarchical data model for supervision and fault detection in the IEC continuum


Monitoring for large infrastructures and applications in heterogeneous IEC continuum environments

  • Current monitoring techniques do not cover the continuum, software stacks and the software-defined network
  • Distributed learning techniques are not adapted to monitoring and anomaly detection in heterogeneous environments

Distributed multi-criteria artificial intelligence for resource use prediction

  • Define a distributed learning model for heterogeneous multi-level infrastructures
  • Define techniques for resource usage prediction
  • Define a software component architecture for resource usage prediction

Applications, validation, implementation and integration

  • Provide real-world application requirements
  • Ensure the correctness and effectiveness of the solutions provided
  • Demonstrate applicability of solutions
  • Provide software components and tools with partial integration in SLICES-FR

— Partners

Consortium

Inria, IMT, CNRS, CEA, Université Versailles
Saint-Quentin en Yvelines, Université de Lorraine

— Research team

13 permanent staff (not funded by the project) :
25 people funded by the project, including:
8 PhD students
8 post-docs
9 engineers

Our teams in France

— Publications



2 documents

Conference papers

  • Tayeb Diab, Mohamed Graiet, Mario Südholt. Declarative, generic definition and effective implementation of transfer learning algorithms. AIAI 2025: 21st International Conference on Artificial Intelligence Applications and Innovations, Jun 2025, Limassol, Cyprus. ⟨hal-05005665⟩
  • Joël Roman Ky, Bertrand Mathieu, Abdelkader Lahmadi, Raouf Boutaba. CATS: Contrastive learning for Anomaly detection in Time Series. 2024 IEEE International Conference on Big Data (Big Data), IEEE, Dec 2024, Washinghton DC, United States. ⟨10.1109/BigData62323.2024.10825476⟩. ⟨hal-04881349⟩


Other projects

close

FORMULAIRE D'ABONNEMENT

Inscrivez-vous pour recevoir chaque trimestre la Newsletter du PEPR Cloud.

En vous abonnant à la newsletter du PEPR Cloud, vous acceptez de recevoir des informations régulières sur nos projets, nos recherches, nos événements, et autres actualités liées au PEPR.

Vous pourrez vous désabonner à tout moment en utilisant le lien de désabonnement présent sur chaque newsletter.

Politique de confidentialité

Comments are closed.