Direct load control has been an efficient and effective load control mechanism under demand side management in smart grid. With this technique, consumers' energy consuming appliances are managed and controlled by central utility with the primary objective of load profiling and to reduce peak load demand. In this paper, we take up the problem of scheduling under peak load constraint for controllable appliances that allows them to be run on various energy consumption levels. The problem is a unique manifold optimization problem where before scheduling a device, we are required to select a specific power consumption level for it that would result in minimizing overall finishing time (delay minimization). We model this problem into two-dimensional bin packing problem and present mathematical programming formulation, while proving it to be NP-hard. We have obtained optimum solutions for 20 numbers of appliances using IBM ILOG CPLEX optimization solver. Further a number of approximation algorithms have been used to obtain solutions quickly.