Optimizing Service Migration and Task Offloading in Osmotic Computing with Deep Q-Networks

Osmotic Computing (OC) has emerged as a system that allows seamless integration of IoT-Edge-Cloud continuum. It aims at optimizing the orchestration and distribution of resources to IoT devices across edge and cloud environments. It enhances service delivery and optimizes response time by dynamically adapting to changing network workloads. This research paper presents a novel method for decision-making in OC environment using Deep Q-Networks (DQN) to intelligently manage and allocate resources. The proposed method has two different operations: service migration and task offloading. Using reinforcement learning, DQN-osmosis learns to make optimal decisions based on current state metrics, such as resource usage. The algorithm adapts to varying network situations and ensures effective resource allocation and enhanced performance. The mathematical model, that forms the basis of the proposed DQN-based approach is also presented. The model is explained through simulations. The results of the model demonstrate enhancements in the speed of task processing and the overall efficiency of the IoT system, proving the effectiveness of DQN in Osmotic Computing paradigm. To validate the hypothesis that the DQN-based framework performs better than other three methods, i.e., Random Agent, Q-Learning, and SARSA algorithm, Wilcoxon signed-rank test is performed. This non-parametric test compares the performance metric of the proposed method with above three algorithms. The results confirm the effectiveness of the proposed method.