Incremental Learning from Multi-level Monitoring Data and Its Application to Component Based Software Engineering

Abstract

Many new Internet of Things (IoT) applications such a disaster early warning systems, video-streaming, automated driving and similar, are increasingly being built by using advanced component based software engineering approaches. Software components can include various executable images, such as container or Virtual Machine images, scripts and others. Achieving adequate Quality of Service (QoS) for such applications is still a challenging issue due to runtime variations in running conditions intrinsic to the cloud, edge and fog environments. These types of systems should therefore be continuously monitored and hence adapted at various levels including infrastructure, container and application levels. In this work, we present an adaptation method using a new Incremental Learning approach based on Multi-Level Monitoring data. The method dynamically generates a set of rules representing a performance prediction model that allow us to find potential performance bottlenecks and then propose suitable application adaptation actions. Adaptation possibilities in this work include (1) live-migration of application components (such as containers) from the current infrastructure to another one with different characteristics, such as CPU, memory, disk or bandwidth capacity, and (2) dynamic horizontal or vertical scaling of container-based application instances to offer better fitted resource capacities.