Scheduling Methods and Models for Kubernetes Orchestrator

In the conditions of evolution of cloud technologies, data centers use Kubernetes orchestrator more and more often, it enables the efficient management of containerized applications. At the same time, Kubernetes is not perfect, and its usage is related to certain problems, among which the problem of efficient scheduling can be highlighted. Its relevance can be explained by the fact that in-built kube-scheduler module does not always build the most efficient schedules. The relevance is intensified by the known cases when inefficiently formed schedule resulted in impossibility to deploy the application. Maximization of coefficient of average workload of the node was chosen as an optimization criterion. It was done based on the objectives of data centers to decrease the energy costs, and the assumptions about inefficiency of computing resources idling. It was determined that mathematically formalized constraints as parts of mathematical models are usually mentioned in the publications dedicated to heuristic methods and metaheuristic methods. In total, six main types of mathematically formalized constraints were determined; the most common and important among them is the memory size constraint. It is highlighted that the method that is getting chosen for solving a problem can be linked with the chosen optimization criterion. Generally, nine main types of methods that are used in problems of effective scheduling for Kubernetes were determined. Among them, three were chosen as the most promising ones: artificial intelligence methods, heuristic methods and metaheuristic methods. The reasons behind their selection include the examples of their successful usage in the formation of schedules with wide spectrum of optimization criteria (including problems with criteria that are similar to the chosen one) both in the cloud environments that use Kubernetes and the cloud environments that don’t use it.