A Proactive Migration Strategy for MEC Applications in 5G-enabled Vehicular Networks

The quality of service for delay-sensitive applications in vehicular networks - such as real-time video streaming - can be greatly improved with the adoption of 5G networks and Multi-access Edge Computing (MEC). MEC allows vehicular applications to be hosted on virtualized infrastructure at the edge of the network rather than in remote cloud data centers, leveraging improved computing power while keeping end-to-end latency low. In order to experience optimal performance, the MEC application should be kept as close as possible to the vehicle, hence it should be migrated to a different MEC host when the vehicle performs handover to a different radio base station. However, application migration takes a non-negligible, longer time than a radio handover, and this may introduce intolerable delays and hinder service continuity because the vehicle should either halt its communications or stay connected to the previous—suboptimal—MEC host until the migration process has completed. To address this issue, we propose a proactive application migration strategy that leverages a sequence-to-sequence Long Short-Term Memory (LSTM) model to predict radio handovers and trigger timely migration in advance. Moreover, we showthat applications based on Transmission Control Protocol (TCP) are most affected by the above problem due to the congestion control mechanism that may mistakenly confuse higher delays due to migration with network congestion and reduce the application throughput. Thus, we propose a mechanism to properly adjust the size of the TCP congestion window after radio handover based on the MEC Radio Network Information Service (RNIS). System-level simulations showthat our proactive application migration solution significantly reduces end-to-end delay compared to traditional reactive strategies, while also preventing unnecessary throughput degradation by optimizing the size of the TCP congestion window.