Analysis, modeling and implementation of cost models for a multi-cloud Kubernetes context

In the last years cloud computing is becoming more and more important for companies. They start developing their applications to be cloud-native, composed by a bunch of containerized micro-services cooperating each other in order to exploit the horizontal scalability capabilities. Having many containers require a component able to manage their life cycle: the widely used solution is Kubernetes. More and more companies are interested in diversifying their cloud infrastructure in two main ways: multi-cloud, that is having clusters at different public cloud providers, or hybrid cloud, an owned physical data center and a virtual infrastructure at one or more cloud providers. Diversifying the infrastructure allows to increase resiliency, to be present on different geographical regions, to avoid lock-in to a single provider and gives the opportunity of saving costs. The last point is what this thesis wants to investigate. How can we exploit that kind of infrastructures optimizing costs? Public cloud clusters are elastic: the number of nodes can increase and decrease in time to adapt their overall capacity to the actual workload in execution. This has to be kept into consideration when modeling their costs. Knowing how Kubernetes cluster costs are structured can be very useful for many reasons: for teams to increase awareness on how much are spending, to compare what cloud provider is more convenient for some requirements, to estimate the costs of having a private datacenter and comparing it to having virtual infrastructures. Cloud providers bills, unfortunately, does not give at all details on how costs are distributed, so having a model can make them more clear. An effort has been made on modeling container costs since they are the smallest execution unit in Kubernetes. Two kind of models have been defined: the overall Kubernetes cluster cost model (one for the public cloud, one for the private cloud) and the container cost model. The knowledge of the container cost model has been then applied to a multi-cluster scheduling scenario to increase the cost efficiency of the container distribution among the clusters thanks to their cost information.