Development of Kubernetes Simulator

Any modern high-load service cannot scale if not based on distributed architecture. Computing cluster for such services may consist from hundreds and event thousands of nodes. With constantly increasing capacity of a cluster more sophisticated cluster management challenges arise. Kubernetes is open-source container orchestration system for automating software deployment, scaling and management which is used by the largest cloud providers such as AWS, Google Cloud, Azure and many other tech companies all over the world. Underlying algorithms of such Kubernetes components as scheduler, cluster and pod autoscalers are on the agenda for researchers today. Often the size and the cost of a cluster make comprehensive research impossible to conduct in reasonable amount of time. Computer model or simulator is used for such purposes. It should simulate the work of the most important components with enough precision, at the same time omitting insignificant details of implementation for fast and simplistic approach. DSLab – is a framework for building discrete-event simulation systems. Any object in a framework is a simulation component which can process events and send them to other components. In this paper the development of Kubernetes cluster simulator based on DSLab is presented. It contains all crucial components and functions of the original software. Simulator introduces a wide variety of useful and convenient extending interfaces for resource management researchers. It is possible to validate research hypotheses fast and cheap on commodity hardware at the scale of 1000 nodes and 100000 pods. The distinctive features in comparison with other simulators are support of trace events processing, high-perfomance and autoscalers modeling.