DQN-Based Applications Offloading with Multiple Interdependent Tasks in Mobile Edge Computing

Recently, Vehicular Edge Computing (VEC) is evolving as a solution for offloading computationally intensive tasks in in-vehicle environments. However, when the number of vehicles and users is large, pure edge resources may be insufficient and limited, most existing work focuses on minimizing system latency by designing some offloading strategies. Therefore, hybrid multilayer edge structures are in dire require of mission deployment strategies that can synthesize cost and mission latency. In this paper, we argue that each application can be decomposed into multiple interdependent subtasks, and that the different subtasks can be deployed separately into different edge layers in a hybrid three-tier edge computing infrastructure for execution. We develop an improved DQN task deployment algorithm based on Lyapunov optimization to jointly optimize the average workflow latency and cost under a long-term cost constraint, and simulation results clearly show that, comparing with the traditional approach, our proposed method effectively reduces the cost consumption by 92.8% while sacrificing only some latency.