ScholarWorks@동국대학교

초록

Container resource autoscaling provides scalability by adjusting the size and number of containers based on the load in container-based fog computing environments. However, fog nodes have limited resources and cannot scale effectively to large-scale loads, making it challenging to ensure service performance. Therefore, an offloading technique is combined with an autoscaling technique to provide service continuity and scalability. However, both techniques operate based on a reactive mechanism, resulting in wasted resources and overloading from dynamic loads. Therefore, we propose efficient proactive resource management (EProRM) to ensure resource efficiency and service continuity in container-based fog computing environments with limited resources. In addition, EProRM independently collects the resource metrics of microservices running on each fog node and predicts the future workload via a decomposition network (DecompNet) model using split learning. Next, we identify burst states in the predicted workload to ensure service stability from dynamic loads. Finally, EProRM performs proactive mechanism-based autoscaling and offloading based on the identified burst states. Moreover, EProRM improves resource utilization by up to 18.33% and reduces the number of instances of overload by about 429 compared to existing resource management techniques.

키워드