Containerized Cloud Platform-Orientated Design And Implementation Of Resource Scheduler

As computation continues to move into the cloud,the computing platform of interest no longer resembles a small rack,but warehouse-full computers.With the rapid development of container techniques,among which Docker owns much reputation,the new choices are showing up for the infrastructure of data centers these days.Since containerized cloud platform is proud of its lightweight and flexible framework,numerous use cases of the industry are highly welcome in the area of computation,storage,network and cloud management.Resource scheduling management plays an extremely important role for data centers.With an efficient resource scheduler,we are able to reduce power usage as much as possible,with the guarantee that we won't suffer service failures.Nowadays most of the schedulers of containerized cloud platform are set up on queue model or distributed two-level scheduling mechanism.On the one hand,we could benefit from its simple and easy-pluggable structure,while on the other hand,this could also be an obstacle for the global optimization attempt.To address this issue,this dissertation adopts a solution that convert the resource scheduling issue into a minimum-cost maximum-flow problem based on the flow network.We proposes a cloud platform orientated cost model called multi-tenant,multi-zone model,which takes advantage of the multi-zone feature in the containerized cloud platform.Also we introduce significant improvement to coordinated co-location cost models,such that we could build up an expressive way for sophisticated scheduling policies.In order to verify the proposed algorithm,we conduct the experiments on Kubernetes cluster.Our results show that this algorithm can achieve good task assignment for containerized cloud platform,and is ready for the potential scalability issue in production environment.