In this talk we discuss possible architectures for 
coordinated home energy management (CoHEM) systems. In our 
model several households use Home Energy Management (HEM) units to schedule collaboratively non-interruptible tasks,
that are requested at random. The aggregate cost for the community is assumed to be an increasing function of deviation between the customers collective electrical consumption and the power supply available to serve them. 
 
We demonstrate that the proposed CoHEM architecture
can effectively improve real-time power balancing, and discuss the trade-offs in accuracy of these architectures compared to one that is computing schedules in a central manner, for a large population.