Predictive Scheduling for Networks: Fundamental Limit, Optimality and Trade-offs


    A natural optimization of online decision making is to consider leveraging recently developed machine learning techniques to predict future traffic information to reduce response time and improve quality-of-service. There is no free lunch, though. For example, Netflix preloads videos onto users’ devices based on user behavior prediction, but such a preloading can be wasteful if wrongly decided. Despite the wide adoption of such prediction-based approaches, it still remains open what are the fundamental benefits of predictive scheduling to netwoks such as fog(edge) computing and cloud computing networks, even in the presence of prediction errors. By developing an innonative framework, our work is the first to answer the questions, which are the key to understand whether the endeavor worthy to put on predictive scheduling, whether one can tolerate the worst possible case that may occur, and various trade-offs shaping the network design.  


    Prof. Ziyu Shao and his collaborators developed a design framework with theoretical analysis and algorithm design to characterize the fundamental limits of benefits of predictive scheduling, robustness of predictive scheduling, and various design trade-offs.