In this paper, the author proposes an Energy-onDemand (EoD) system based on combinatorial optimization of appliance operation modes, and describes its implementation and evaluation. Recently, efficient usage of limited amount of electrical energy has been an important issue. EoD is a novel power network architecture of demand-side power management, whose objective is to intelligently manage power flows among power generations under the limitation of available power resource. In an EoD system, the importance of each appliance is explicitly parameterized, and the amount of power consumption of appliances is measured by power sensors. When total power consumption exceeds the limit of power resource, a power allocation manager deployed in the system decides the optimal power allocation to all the appliances based on their parameters, and controls the amount of power supplied to the appliances in a way that causes minimum undesired effect to quality-oflife (QoL) of users. Therefore, one of the most crucial factors in an EoD system is the strategy for deciding the optimal power allocation. From a mathematical viewpoint, the power allocation management in an EoD system can be considered as an optimization problem of appliance operation modes. In the developed system, power allocation is based on the multiple-choice knapsack problem (MCKP), a kind of combinatorial optimization problem. The system measures power consumption of appliances, computes the optimal power allocation based on an algorithm for the MCKP, and realizes computed power allocation by controlling IR-controllable appliances and mechanical relays.