Balancing Accuracy-Latency Tradeoff in Hybrid Approximate-Delayed Computing Systems

Distributed edge computing, empowering Compute First Networking (CFN), can be enhanced with additional computing paradigms to further minimize latency and ameliorate task processing through edge-cloud collaboration. This paper envisions hybrid approximate-delayed computing across the edge-cloud continuum to optimize task processing efficiency for user-owned consumer devices. In this context, we study the joint problem of task offloading to the edge-cloud system and task scheduling at the edge. The users have two offloading options: approximate computing at the edge or delayed but accurate processing in the cloud. The objective is to balance task accuracy and latency among the two computing options. To this end, a fully distributed solution concept is introduced, where the users and the edge system interact hierarchically to determine the task offloading and scheduling, respectively. On the one hand, the users shape their edge-cloud offloading decisions autonomously by participating in a Satisfaction Game between them, targeting to meet a minimum accuracy-latency tradeoff. On the other hand, the edge system optimizes the scheduling of the tasks intended for approximate processing using a Genetic Algorithm. An iterative process between the users and the edge unfolds to conclude the equilibrium point of the hierarchical game between the users and the edge, referred to as SatisfOptimizing Game. Numerical results show the effectiveness of the proposed solution in optimizing resource utilization across the edge-cloud system, guaranteeing user accuracy-latency requirement satisfaction and edge system energy consumption minimization.