Thesis - Efficient Fault-Tolerant Offloading Mechanisms in Edge Networks

Abstract

Edge networks are revolutionizing computing by bringing storage and computational power closer to users, enabling applications such as big data analysis and location-based services. These networks, composed of small base stations called Cloudlets, facilitate efficient computation offloading to reduce latency. However, existing techniques lack robust decentralized strategies to address user mobility, task dependencies, and edge node failures.

This research tackles these critical challenges in edge computing by focusing on three primary issues:

Implementing fault-tolerant offloading algorithms in distributed environments. Accommodating dynamic user mobility. Managing both independent and dependent tasks modeled as Directed Acyclic Graphs (DAGs). Our proposed fault-tolerant offloading algorithms for mobile user devices are designed to handle task interdependencies, ensuring that dependent tasks execute only after their predecessors have completed. The problem is formalized as a constraint satisfaction problem, and optimization models are developed for both independent and dependent tasks, focusing on minimizing task completion latency while maintaining system reliability.

Simulation results using the ns-3 platform demonstrate that our algorithms achieve up to a 60% reduction in task completion time for independent tasks and 65% for dependent tasks compared to baseline methods. The algorithms maintain a 100% task completion rate at user speeds of 0.5, 1.0, 1.5, and 2.0 m/s, although performance declines at higher mobility speeds, suggesting areas for further optimization. To validate the real-world effectiveness of our approach, we also evaluated the algorithms using the London Trajectories dataset, demonstrating their robustness in realistic urban mobility scenarios.

These results underscore the effectiveness of our decentralized fault-tolerant strategies, enabling reliable task offloading under challenging edge computing conditions. This work significantly advances edge computing and sets a benchmark for future research in the field.