Deep Learning-Based Dispatching Conflict Avoider in MEC Systems

Multi-access Edge Computing (MEC) is a promising computing paradigm that brings cloud services closer to mobile devices. In contrast to cloud computing, MEC dispatches tasks to edge servers, offering lower transmission/serving latency. Distributed dispatch algorithms are commonly employed in large-scale MEC systems to reduce information-sharing latency. However, even in small local area networks (LAN), the Round-Trip Time (RTT) required for information sharing is in the millisecond range. When multiple devices generate tasks within a single RTT, similar environmental states can result in identical dispatching decisions, leading to a sudden surge in the load on a specific server. This phenomenon is known as a dispatching conflict. To address this, we propose DCA (Dispatching Conflicts Avoider), a distributed intelligent agent designed to mitigate dispatching conflicts by dynamically regulating the task pools of edge servers in real-time. Simulation results demonstrate that DCA significantly reduces the weighted response time (WRT) compared to baseline distributed heuristic algorithms, particularly under high-concurrency scenarios.