Mydep

Περιγραφή

Kubernetes is an extensible open-source orchestration platform for automating software deployment, scaling, and management of containerized workloads and services. Kubernetes is considered to be the market-leader and the standardized means of orchestrating containers and deploying distributed applications. These applications are often intertwined with a plethora of demanding Quality of Service (QoS) requirements. The process of optimizing the Task Offloading decisions can be classified into three categories depending on the underlying modus operandi: optimizing performance, minimizing energy consumption, and minimizing system cost. At present, the most frequently used Task Offloading policies tend to be based on rather traditional approaches such as random-based, least-load, and  round-robin. These traditional policies are not capable of facilitating QoS requirements. Kubernetes, in particular, leverages load-balancing techniques that are based on the round-robin policy. The round-robin policy belongs to none of the aforementioned categories but instead chooses to distribute the computational load among the available resources in a fair manner that does not guarantee any form of optimality and ability to facilitate the various QoS requirements. The objective of this thesis is to examine various Deep Reinforcement Learning solutions in the context of dynamic task offloading in Kubernetes Clusters, in a manner that is aligned with the aforementioned QoS requirements.

Στοιχεία Εργασίας