Automatic Resource Management in Geo-Distributed Multi-Cluster Environments

Geo-distributed computing environments such as hybrid cloud, multi-cloud and Fog Computing need to be managed au-
tonomously at large scales to improve resource utilization, maximize performance, and save costs. However, resource management in these geo-distributed computing environments is difficult due to wide geographical distributions, poor network conditions, heterogeneity of resources, and limited capacity. In this thesis, we address some of the resource management challenges using container technology. First, we present an experimental analysis of autoscaling in Kubernetes clusters at the container and Virtual Machine levels. Second, we propose a proportional controller to dynamically improve the stability of geo-distributed deployments at runtime in Kubernetes Federations. Finally, we
develop a container orchestration framework for geo-distributed environments that offers policy-rich placement, autoscaling, bursting, network routing, and dynamic resource provisioning capabilities.

Information : Attendance limited to 15 people including jury members - please inform Angélique Jarnoux (Angelique [*] Jarnouxinria [*] fr) if you wish to attend in person.