A Decentralized Multi-Agent Reinforcement Learning Framework for Joint Task Offloading and Service Caching in Fog Computing: A Methodological Approach

The rapid proliferation of Internet of Things (IoT) devices and latency-sensitive applications has significantly increased computational demands at the network edge. Traditional cloud-centric architecture often fails to meet strict delay requirements due to centralized processing and network congestion. Mobile Edge Computing (MEC) addresses this challenge by deploying computing and storage resources closer to end users. However, efficiently managing limited edge resources under dynamic workloads remains a major challenge, particularly when coordinating task offloading and service caching decisions. This work proposes a decentralized multi-agent reinforcement learning (MARL) framework for joint task offloading and service caching in a cloud-assisted MEC environment. Each MEC-enabled base station operates as an autonomous learning agent that observes local system states and dynamically determines optimal offloading and caching actions. The decision process is modeled as a decentralized partially observable Markov decision process to enable scalable distributed learning. The proposed framework aims to minimize overall operational cost by reducing latency, energy consumption, and cache miss rates while improving resource utilization across fog nodes.