Adaptive Application Deployment for Multi-Access Edge Computing Based on Mobility Prediction

In this paper, the proactive application deployment based on mobility prediction for multi-access edge computing (MEC) is studied. Since mobility prediction is commonly imperfect, there is an inherent conflict between the prediction accuracy and the prediction duration (i.e., the length of time ahead). As a result, proactively deploying applications for MEC based on a shorter (longer) term mobility prediction may lead to a higher (lower) accuracy, and thus reduces (increases) the service delay while suffers (avoids) from a larger deployment cost. To strike such balance, we propose a novel adaptive application deployment scheme, taking the mobility predictions of different mobile users in multiple future time periods as the input, for optimizing their corresponding application deployments (i.e., which applications should be deployed on which edge nodes and how long they should be deployed in advance). Specifically, a residual LSTM framework is utilized for mobility prediction, and based on this, a low-complexity greedy algorithm is developed. Simulation shows the feasibility of the proposed scheme and demonstrate its superiority over counterparts.