Achieving optimal admission control with dynamic scheduling in energy constrained network systems

This paper considers optimization of time average admission rate in an energy-constrained network system with multiple classes of data flows. The system operates regularly over time intervals called frames, while each frame begins with a fixed-length active period and ends with a variable-length idle period. At the beginning of the frame, the system chooses a service mode from a collection of options that affect the class and the amount of data flow served as well as the energy incurred in the active period. After service, the system chooses an amount of time to remain idle. The optimization goal is to make decisions over time that maximizes a weighted sum of admitted data rates subject to constraints on queue stability and energy expenditure. However, conventional solutions suffer from a curse of dimensionality for systems with large state space. Therefore, using a generalized Lyapunov optimization technique, we design a new online control algorithm that solves the problem. The algorithm can push time average admission rate close to optimal, with a corresponding tradeoff in average queue backlog. Remarkably, it does not require any knowledge of the data arrival rates and is provably optimal.