ScholarWorks@동국대학교

초록

A container autoscaling technique reduces resource costs in cloud-native computing-based digital twin environments by adjusting the container instances of virtual entities. However, virtual entities generate irregular loads due to diverse operating conditions and anomaly events, making it challenging to manage using the existing reactive mechanisms. Proactive mechanisms rely on the workload forecaster performance, causing inefficient resource management and potential service disruptions due to inaccurate predictions. Therefore, this paper proposes a dynamic resource management scheme, a hybrid mechanism-based container autoscaling technique, to enhance the resource efficiency and continuity of virtual entities in a cloud-native computing- based digital twin environment. The dynamic resource management scheme collects historical data via a reactive mechanism and trains the LightTS model to predict future workloads. Furthermore, the proposed technique dynamically selects the container autoscaling mechanism based on the prediction accuracy of the LightTS model, efficiently and stably managing the resources of virtual entities. Offline simulations in a Kubernetes-based digital twin environment demonstrated that the dynamic resource management scheme improved resource elasticity by up to 16.1 times compared to existing techniques and reduced resource costs and overload occurrences by 18.13% and 97.37%.

키워드