Cost-aware routing for computation offloading in knowledge-defined AIoT

Edge computing plays a crucial role in supporting high-bandwidth and latency-sensitive applications in the Artificial Intelligence of Things (AIoT). These applications often demand both computing and network resources within strict time constraints, yet existing approaches often fall short in jointly considering dynamic destination-path combinations, pricing incentives, and differentiated computation costs. In this paper, we propose a Knowledge-Defined AIoT-based framework that incorporates a cost-aware routing algorithm called CompuRoute for computation offloading. This framework enables collaborative data collection and centralized data aggregation and analysis, supporting efficient cost estimation. Based on the estimated cost, CompuRoute integrates a reverse auction mechanism for selecting candidate edge servers. Next, CompuRoute considers networking states and introduces a multipath routing algorithm based on network flow theory to determine the destination edge servers and routing paths. Experimental results demonstrate that CompuRoute can improve the task success rate and reduce task completion time compared to baseline algorithms, exhibiting scalability across various network topologies.