Abstract:Task offloading is a core component of the mobile edge computing (MEC) paradigm, and its reliability is directly related to the system security and service quality. Once a task offloading failure occurs, it may result in severe consequences. To address the challenges of insufficient task offloading reliability and lack of differentiated reliability assurance mechanisms for various types of tasks, this study first classifies tasks into four criticality levels according to their sensitivity to reliability and latency. An offloading reliability model integrating MEC node failures, channel interruptions, and latency constraints is built. Then, based on the task criticality levels and reliability model, a reliability penalty function based on task levels is designed to impose differentiated constraints on tasks that fail to meet their reliability targets. On the basis of the reliability penalty function, a comprehensive cost model incorporating system energy consumption and latency is built. Finally, the greedy-based hybrid simulated annealing for task offloading optimization algorithm (GHSA) is adopted to optimize and solve the comprehensive cost. Under different experimental scenarios, the performance of GHSA is compared with that of task offloading algorithms, including BTOA, AGA, SA, and RA. The simulation results show that compared with the comparative algorithms, GHSA increases the total task offloading success rate by an average of 21.43%–48.17%. For high-criticality tasks, the average improvement in the offloading success rate exceeds 33.60%. The results fully verify that GHSA can effectively balance system energy consumption and latency while ensuring reliable offloading of high-criticality tasks, showing strong overall optimization capability and adaptability.