Intelligent and Secure Dependency-Aware Partial Task Offloading in IoT Edge Networks

Abstract

The rapid growth of the Internet of Things (IoT) has driven the need for low-latency processing, which is not readily provided by traditional cloud architectures. Multi-access Edge Computing (MEC) can address this problem by relocating resources to the network edge, but it faces challenges in handling complex subtask dependencies, network dynamism, and security vulnerabilities. Although a body of research exists on combinations of different technologies to tackle this problem, one gap remains: no single model combines Blockchain, Deep Reinforcement Learning (DRL), Software Defined Networking (SDN), dependency awareness, and partial offloading. This thesis proposes an intelligent and secure system to fill this gap by integrating these technologies into a single system. The architecture is based on an SDN controller that provides a global view of the network and manages flow control, a private Ethereum blockchain to access resources securely via a token-gating mechanism, and a Deep Q-Network (DQN) agent to optimise partial offloading decisions for subtasks modelled as Directed Acyclic Graphs (DAGs). The experimental results show that the DQN-integrated methods (DQN and DQN+BC) are more efficient than the heuristic baselines, reducing the average task completion time by 78–85%. Moreover, the standalone DQN system achieves a 100% task completion rate across varying workloads, whereas heuristic approaches often discard tasks due to time-constraint violations or resource limitations. The DQN+Blockchain strategy also maintains a 100% task completion rate in most cases, except at high workloads, where the blockchain gating mechanism actively restricts the operation of devices that consume excessive amounts of edge resources, demonstrating the intended role of the blockchain in preventing uncontrolled and abusive use of the shared MEC infrastructure. Although adding blockchain incurs a slight 5-10% increase in completion time, it provides security, accountability, and traceability for the use of edge resources. The work provides a strong foundation for the development of scalable, secure, and efficient intelligent edge systems.