Deep Reinforcement Learning for Intelligent Migration of Fog Services in Smart Cities

Fog computing plays a crucial role in future smart city applications, enabling services running along the cloud-to-thing continuum with low latency and high quality of service (QoS) requirements. However, the mobility of end users in smart city systems can result in considerable network performance and QoS degradation, hence interrupting fog services provisioning. Service migration is considered an effective solution to avoid service interruption and ensure service continuity, which can be carried out proactively or reactively. Existing work lacks intelligent and efficient migration solutions for fog services migrations. In this paper, we propose Octofog, a fog services migration model and framework in the context of smart cities, featuring artificial intelligence for resource-efficient migration. We formulate proactive and reactive migration policies as an optimization problem, minimizing migration cost in terms of delay and energy consumption. We use a deep reinforcement learning (DRL) algorithm to solve the optimization problem to make fast migration decisions, using deep deterministic policy gradient (DDPG) based schemes. The evaluation results illustrate that Octofog effectively reduces the total migration cost (i.e., latency and energy).