Dynamic load balancing in 5G HetNets for optimal performance-energy tradeoff

We consider optimal energy-aware load balancing of elastic downlink data traffic inside a macrocell with multiple small cells within its coverage area. The model for the problem corresponds to a system of parallel M/M/1-PS queues, where the macrocell is represented by a multiclass M/M/1-PS queue and each small cell is an energy-aware M/M/1-PS queue with additional states for the idle timer and the so-called setup delay. We apply the theory of MDPs to develop a near-optimal state-dependent policy, both for a weighted sum of the performance and energy as well as for the constrained formulation, where energy is minimized subject to a constraint on the performance. Specifically, we utilize the first step of the well-known policy iteration method under which the routing decision for each arrival requires evaluating the marginal future cost of adding the arrival in the small cell or the macrocell. As our main contribution, we derive the associated value functions and the explicit form of the near-optimal FPI policy. The performance of the policy is illustrated through numerical examples.