Abstract: Lightweight virtualization, and specifically containers, has become widespread in the information technology industry to provide an efficient operational environment for the execution of scalable services on the Internet. Containers rely on a set of technologies different from the features that enable hardware virtualization (i.e., hypervisor-based virtual machines). However, both types of virtualized environments are employed to host applications that will be accessible anytime, anywhere. Thus, they are prone to software aging, which usually affects systems that run for long time intervals. Researchers have identified software aging effects in distinct types of cloud computing environments and hypervisors over recent years. However, a few works have dealt with this phenomenon in container-based platforms. This paper presents an experimental analysis of the software aging effects observed on Docker platforms while also evaluating the fitness of a time-series model to predict resource consumption’s progress caused by software aging. We employ a stress test workload tailored for the scenario of containers arranged in a cluster managed by Docker Swarm. The obtained results indicate an increasing usage of resident memory, virtual memory, and CPU usage, as the system undergoes subsequent scale-up and scale-down operations. The quadratic trend model was the best fitting approach for predicting resident and virtual memory usage, with less than 5% of prediction error. The experimental approach presented here may help systems administrators to detect evidence of software aging in container-based environments, and allowing then to choose a proper method and time for deploying rejuvenation actions to mitigate the dependability issues raised in similar scenarios described here.
Details in this link.