.png){kind=icon}
Return
Task-driven priority-aware computation offloading with distributed CPU resource allocation in edge-cloud IoT networks
.png){kind=icon}
PRE
AI
DOI：10.1016/j.iot.2026.102068.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Efficient computation offloading is essential in IoT edge-cloud networks, where resource-constrained devices must process heterogeneous and delay-sensitive workloads. Most existing approaches rely on slot-based decision making and fail to capture the task-driven and asynchronous nature of practical IoT environments, resulting in increased decision latency and reduced responsiveness for high-priority tasks. In this paper, we propose a task-driven and priority-aware computation offloading framework where decisions are triggered immediately upon task arrival. A deep reinforcement learning agent based on prioritized experience replay dueling double deep Q-learning (PER-D3QN) is employed to jointly determine the execution location and transmission power under dynamic channel conditions and queue states. To manage contention among concurrently offloaded tasks, a distributed CPU sharing mechanism is executed locally at each edge server, enabling adaptive allocation of computational resources according to task workload and priority. The resulting execution delay and queue evolution are treated as part of the environment dynamics and provide feedback to the learning agent. Simulation results demonstrate that the proposed framework significantly reduces task execution delay and IoT device energy consumption while improving deadline satisfaction and load balance compared with conventional DRL-based and heuristic offloading schemes.
Keywords:
Task-driven offloading
Deep reinforcement learning
Dueling double deep Q-network
Distributed CPU sharing
Internet of things (IoT)
Journal
IF:
7.6
Papers:
1.9K
Citations:
6.9K
Organization
Cited Papers
No cited papers available