In recent years, the remote sensing infrastructure is slowly shifting towards satellite constellation systems from monolithic satellites. A constellation offers high-fidelity (both in time and space) distributed sensing capability. The prospect of using a small satellite constellation as a commercial infrastructure for serving a multitude of applications for multiple users, is quickly becoming a reality. However, such infrastructures demand handling of high volume of sensor data with high velocity. In addition, commercial exploitation also requires capability to handle variable sensing demands depending on user requirements. Therefore, it is a challenging problem to efficiently utilize on-board sensing resources to make the infrastructure commercially attractive. In this paper, we present the sensing task (aka collection) scheduling problem under limited storage constraints in satellites. We present an integer linear programming (ILP) based solution and show that it produces higher throughput in comparison to state-of-art collection scheduling methods, through extensive simulation. Further, we present a linear relaxation and randomized rounding based heuristic, with guaranteed approximation bounds, and show experimentally that it is comparable with the optimal solution.