Based on measurements obtained by commodity wireless sensors, we observe that the majority of thermostatic loads in a user premise are described by periodic pulse waves. We propose a novel first stage of optimization in the smart grid which reduces external on/off command flow for demand response between the controller and the smart appliances. A phase management scheme is developed that defines optimal time shifts on the periodic loads in order to provide peak power and energy cost reduction over a limited time horizon. A gradient descent optimization technique, based on Taylor series, is applied to determine the phases of the pulses in discrete time steps. Three optimization strategies and two control schemes are explored. Minimization of peak power loads, minimization of energy costs and flattening of the power curve are modeled. A centralized and a distributed control scheme are explored. It is found that respectable peak power and cost reduction can be achieved in the centralized control scheme but redundant data transfer in the network and increased complexity is necessary. On the other hand, the distributed control scheme reduces the overall complexity but does not present significant savings.