Résumé: Nowadays, cloud computing faces growing challenges, furthermore, responding to time-sensitive requests in the traditional cloud computing model is one of the major challenges, considering the growth of the Internet of Things. Recently, the powerful fog computing paradigm has been considered for answering these challenges. But a major challenge in fog computing is managing limited FN resources for correctly responding to IoT requests in environments with heterogeneous latency requirements. This study presented a new resource management framework utilizing a software defined network (SDN) architecture and enhanced reinforcement learning methods. This new framework aims to make optimal use of limited FN resources while satisfying the low-latency requirements of IoT applications. Since SDN is the proper choice to support this intelligent distributed structure, an SDN-based fog architecture was proposed. Moreover, FN must allocate its limited and valuable resources intelligently in a heterogeneous IoT environment with different latency requirements. Consequently, this study formulated the problem of resource allocation in fog computing in the form of a Markov decision process (MDP). The study used different reinforcement learning (RL) techniques to solve the MDP problem. Simulation results in different IoT environments with heterogeneous latency requirements corroborate that RL methods achieve the best possible performance regardless of the IoT environment.

Langue: Anglais