Efficient Container Management Scheme Based on Deep Learning Model

The container orchestration platform provides services and applications to users by automatically managing containers in cloud-native clusters. However, the conventional container management technique operates based on a reactive mechanism, so it is difficult to guarantee availability for a rapidly changing workload, and a problem arises in which resources are wasted. Therefore, this study proposes an efficient container management scheme (ECMS) that guarantees high availability and scalability of cloud services and applications composed of various workloads and minimizes idle resources. ECMS predicts future workload by training accumulated past resource usage metrics on a deep learning model. ECMS creates a container by scaling the size of the resource to prevent the generation of idle resources, then selects a node that guarantees high availability and deploys the container. Lastly, it provides scalability of service by performing horizontal resource container autoscaling based on proactive mechanism.