Capacity-Based Trust System in Untrusted MEC Environments

Abstract

With the concept of Mobile Edge Computing or Multi-access Edge Computing (MEC), computation resources can be located closer to the users in proximity. MEC can operate specific services for the user requests to reduce overhead in the main cloud. The MEC infrastructure also facilitates service migration and relocation to follow the user movement in order to maintain service connection and therefore reduce the service downtime. However, since most MEC components are meant to be deployed on constrained IoT devices, they inherit many of the limitations that come with IoT, especially in terms of limited resources and security. In this paper, we focus on the capacity aspect of the MEC infrastructure to ensure the services have enough resources during the migration process. We note that the distributed and decentralized nature of MEC networks requires the additional endorsement of the capacity information reported by the network nodes, which is especially critical since MEC nodes can be hacked or tampered with, leading to failures during the migration process. To address this issue, we propose a machine learning-based solution that applies historical resources metric data to predict potential node behavior under attack scenarios. Our experiments showed that artificial neural network (ANN) is the best model for the production version that will be built as an extension in Kubernetes.