Scheduling Jobs on Federation of Kubernetes Clusters

In the last two decades the cloud has gained a lot of importance, indeed the current trend is to engineer the new web applications to be cloud native, thus to be split up in loosely-coupled micro-services, each one containerized and deployed as a part of a bigger application. The use of containers allows to cut oneself off the hosting physical hardware and operating system, letting to focus on the main purposes of a web application: to be widespread and high-available. The cloud allows to achieve this goal, by gathering the infrastructure control under the cloud provider tenants and implementing the IaaS (Infrastructure as a Service) and PaaS (Platform as a Service) paradigms: the computational, networking and storage resources are provided on demand to the cloud provider's customers as if they were services. A technology that broke through the cloud market is \kubernetes, a project kicked off by Google in 2014 that allows to automate deployment, scaling, and management of containerized applications. Beside the cloud, in recent years a new paradigm has gained a lot of importance, the edge computing: a distributed computing paradigm that brings the computational and storage resources close to the final user, in order to improve the QoS standards in terms of latency and bandwidth. The goal of the project in which this thesis is involved is to create a federation of \kubernetes{} clusters that cooperate at the network edge: many different tenants are connected together to cooperate in creating a federation of clusters with computational, storage and networking resources shared between them. In this scenario every tenant can make own resource cluster available to the federation by sharing or leasing them out in a federated and democratic environment. This model needs a scheduler able to take into accounts both computational and inter-connectivity requirements for performing a \scheduling{} that satisfies QoS policies set by the application. The proposed scheduling algorithm makes use of an heuristic approach to pair every job to be deployed to a cluster of the federation: a deployment composed by a set of jobs (i.e. set of micro-services), each one with different constraints in terms of computation, memory, storage and inter-connection is represented by a graph that is scheduled on the graph of federated clusters. The function to be optimized by the algorithm is the cost for the leasing of the set of jobs.