Design of Reactive Systems for Control Network Traffic on the Kubernetes Platform

Lubos Mercl,Vladimir Sobeslav,Peter Mikulecky

DOI: https://doi.org/10.1007/978-3-030-14132-5_33

2019-03-06

Abstract:The container virtualization on the Mercl, LubosSobeslav, VladimirMikulecky, Peter challenges that need to be addressed when a production or test load is running across the cluster. If applications running in containers are spread across the Kubernetes cluster, resource usage may be inefficient across the environment, which may result in overloading of individual nodes and inefficient load on others. One area where inefficiencies may occur is the load on network lines and network communications. This article discusses two algorithms and approaches that can be applied to the Kubernetes platform while helping to manage network traffic and lines across the cluster, which can make the cluster components more efficient. Both algorithms collect the monitored data from the cluster, but each one the data collected behaves differently, and data transformation and analysis takes place in another part of the system. The first algorithm is an agent-based algorithm that collects and performs basic data analysis and is capable of responding to detected information. The second is the algorithm that collects the data into the central element of the system and then analyzes it and, based on the information obtained, controls the individual components of the cluster.