We consider in this paper how the increasingly popular broadband access can be leveraged for scalable and cost-effective video service. Although much work has been done for such service, the existing techniques either are not designed for on-demand video delivery, or do not take into account the vastly improved client receiving capability. In this paper, we address this problem and develop three novel video delivery techniques with two ultimate goals: minimizing service latency and maximize data sharing. With the new techniques, a client can be served as soon as server resource becomes available and meanwhile, all its receiving bandwidth can be used for data receiving. In particular, in the proposed dynamic scheduling algorithm, the efficiency of data sharing is further improved by also exploring server bandwidth for early delivery of video segments. This innovation is contradicted to the intuition that each video segment should be delivered as late as possible in order to maximize data sharing. We evaluate the performance of our new techniques using simulation and our study convincingly shows that with the new schemes, zero service latency can indeed be achieved with very minimal system resource.