Abstract
In cloud computing, resizing component resources is often limited by the available resources offered by a provider. After reaching a resource limit, a component cannot acquire more resources, which can badly affect the load situation. This article presents multiple predictable recovery actions of a self-healing model for an identified anomalous behavior (eg, overload, underload) to auto-scale compute resources in a containerized cluster environment according to various workload conditions. The efficacy of the model is demonstrated through an evaluation with different auto-scaling strategies based on the number of created/terminated containers, container migration, resource utilization, and response time. The results show that the proposed model provides promising overall performance under dynamic workloads compared to other auto-scaling strategies.