Investigação dos efeitos do envelhecimento do software e avaliação de políticas de rejuvenescimento em sistema de orquestração de conttêineres

Abstract

Software Aging and Rejuvenation (SAR) have garnered attention in recent years and have been extensively studied in various computational scenarios, especially in systems requiring high availability. This focus reflects the increasing importance of software systems for the efficient operation of various organizations, becoming a prominent research topic driven by the growing demand for reliability and service availability. This dissertation presents experiments conducted on two versions of Ubuntu Linux, simulating operation scenarios of a private cloud. The containerization platform used is Containerd, with Kubernetes (K8s) as the orchestrator, in 5 distinct scenarios. For each of the 5 scenarios, hardware and software conditions were maintained, including workloads and test durations. During the experiments, metrics such as CPU usage, memory usage, and disk usage were monitored throughout the system, in addition to analyzing the services of Containerd and Kubelet. The experiments also included measuring the web server response time for external HTTP requests. The initial scenario focused on investigating the effects of software aging, while subsequent scenarios explored the adoption of different rejuvenation strategies. Effects of software aging were observed in all scenarios, with resource leaks identified, especially in memory usage, even when the cluster was idle after a stress period. The issues observed in this work result in system performance degradation and compromise reliability and availability when the system fails due to memory usage exhaustion. Rejuvenation had no effect in scenarios involving memory cache cleaning, let alone during the restart of Containerd and Kubelet services. However, software rejuvenation proved effective in the final scenario when controlled restarts of Worker nodes were performed, thus allowing compliance with the stipulated experiment time.