DIMO: Constrained Computing Power Information Dissemination with Multi-hop Task Offloading

The proliferation of Internet of Things (IoT) applications is constrained by the limited computing power of IoT devices, which can be effectively addressed by Multi-access Edge Computing (MEC) through multi-hop task offloading. This paper addresses a key problem of how to disseminate the computing power information (CPI) of edge servers and support the IoT devices to select appropriate edge servers at acceptable CPI dissemination cost. First, the CPI dissemination cost is derived with respect to the dissemination distance based on a tree. Then, the impact of CPI dissemination distance on the multi-hop task offloading is studied, and the corresponding task offloading delay is obtained by employing queuing theory. On this basis, we propose DIMO, a constrained CPI dissemination scheme, to minimize the CPI dissemination cost and average task offloading delay. Specifically, DIMO employs a deep reinforcement learning framework to determine the optimal dissemination distance of each server, wherein a gradual adjustment strategy and proximal policy optimization (PPO) algorithm are utilized. Experimental results demonstrate that DIMO can reduce the CPI dissemination cost by 66.88-95.14%, while maintaining similar task offloading delay, compared with the current state-of-the-art. Furthermore, DIMO can reduce the overall cost by up to 66.17% to 82.37%, achieving an optimal balance between CPI dissemination cost and task offloading delay.