SIA: A Stackelberg-induced Actor-Critic framework for optimized cloud infrastructure scheduling

Abstract

An innovative Stackelberg-induced Actor-Critic Framework (SIA) is proposed to address QoS optimization and load balancing challenges in cloud infrastructure. SIA models their interaction as a hierarchical control problem using a Stackelberg game, where the leader (QoS components) anticipates the follower's (load balancing components) responses; both influence task assignments to virtual machines. Operating on a discrete-time nonlinear system, SIA employs adaptive critic design with neural networks. Experimental results show that SIA substantially improves QoS prediction in cloud workflow scheduling, achieving RMSE reductions from 22.93% to 35.82% and MAE reductions exceeding 100% compared to existing methods.