Towards 6G Joint HAPS-MEC-Cloud 3C Resource Allocation for Delay-Aware Computation Offloading

Abstract

The past years witnessed the tremendous growth of Internet of things (IoT) service, each of which demands different amounts of physical resources consisting of computation, communication, and caching, which is also recognized as 3C. The fifth-generation (5G) technique is a promising answer to serve delay-sensitive IoT applications with diverse popular emerging techniques such as multi-access edge computing (MEC) and cloud computing. However, when we get to 2030, the requirements will hard to satisfied. The sixth-generation (6G) aims to provide global coverage, enhanced energy and cost efficiency, better intelligence level, and security. A potential solution for the 6G system is the aerial access network (AAN). The high altitude platform system (HAPS) is also a candidate for deploying wireless communications applying the terrestrial communication infrastructure. However, how to efficiently utilize the 3C resources in the HAPS-terrestrial networks is a non-trivial issue. We study the offloading computation problem of the IoT applications which ask 3C resources in the HAPS-MEC-cloud networks with high efficiency. In detail, We formulate the computation offloading problem into an optimization problem to minimize costs under multiple resource constraints. Since the problem is integer linear programming (ILP), it is hard to apply the general exhaustive searching to solve the problem when there are a lot of mobile terminal devices. The column generation algorithm can solve the large-scale ILP problem efficiently. Thus, we propose a column generation computation offloading (CG-CO) algorithm based on it. Meanwhile, we proposed a greedy computation offloading algorithm (G-CO) based on a greedy algorithm for comparison to make the simulation results in more convictive. We use the task acceptance ratio, service providers' total revenue, and cost as the metrics. Experiment results demonstrate that CG-CO can help get good results in both resource-abundant and resource-limited scenarios.