Dependent Task Offloading and Dynamic Load Balancing in Edge Computing Using Deep Reinforcement Learning With Consistent Hash

Multi-access edge computing is widely used to solve the problem of delay and energy consumption when deploying compute-intensive applications on user devices at the network edge. Optimizing offloading strategy for dependent tasks of user devices and load balancing of edge servers is crucial for multi-access edge computing. However, due to the limited computational resources of edge servers, reasonable offloading and allocation for dependent tasks is a great challenge in multi-access edge computing. To address this problem, this paper proposes an optimization approach for dependent task offloading and dynamic load balancing in edge computing using multi-agent soft actor-critic. The proposed approach designs the computing model and dependent task model by taking dependent task operational constraints and dynamic load balancing into account. Then the offloading strategy of dependent tasks is optimized to get lower delays and energy consumption by the multi-agent soft actor-critic integration with a kind of consistent hash-based load balancing mechanism. Simulation results show that the proposed approach has obvious advantages in terms of energy consumption of user devices, execution delay of tasks, and load balancing of edge servers.