A fine-grained traffic engineering (TE) that enables per-flow control is considered to be necessary in future Internet. In this paper, we study to realize flow-level TE in conventional networks, where hop-by-hop routing is available, and advanced technologies such as SDN and MPLS are not deployed. Based on analysis and modelling on real Internet traffic, we propose to detect and schedule a few large flows in real time, which dominate the traffic amount. The proposed scheme leverages advanced algorithms for detection, computes the rerouting paths in a centralized server, uses extended OSPF to distribute the routing, and uses a few ACL entries for flow-level forwarding. We formalize the link weight assignment-based large flow scheduling problem and prove that the problem is NP-hard. We develop algorithms to compute the routing and reduce extra LSA number required. We present a set of theoretical results on the TE performance bounds when the number of large flows varies. Experiment and simulation results show that our scheme can reroute large flows within 0.5 second, and the maximum link utilization is within 102% of the optimal solution for source and destination addresses-based flows, while the extra LSA number is small.