Self-Management of Containers Deployment in Decentralized Environments
Fabiana Rossi · 2019
The increasing presence of IoT and fog computing resources are the changing computing environments, leading the applications towards an elastic and decentralized execution. To simplify the deployment and management of applications, containers are widely adopted nowadays. However, although several solutions for orchestrating containers exist, most of them are not suitable to operate in the emerging computing environment, which exposes new challenges due to heterogeneity of computing resources and dynamism of working conditions. The doctorate work, presented in this paper, investigates the run-time adaptation of containerized applications with Quality of Service requirements, deployed over heterogeneous computing and networking resources. We propose decentralized policies based on reinforcement learning (RL) to adapt the application deployment, in terms of elasticity and container migrations. As first research step, we design and evaluate RL solutions that exploit different degrees of knowledge about the system dynamics (i.e., Q-learning, Dyna-Q, and Model-based). As future research, we want to design more efficient adaptation heuristics, exploiting the possibilities of IoT and fog environments, and considering mobility of users and resources within the adaptation loop. Furthermore, we plan to investigate multi-level elasticity, that jointly optimizes applications and infrastructure elasticity.