In this paper, we present a predictive congestion control policy for transporting data traffic in a high-speed wide-area network, in the presence of real-time traffic, such as voice and video traffic. The policy is based on information exchange between adjacent nodes of the network. The evolution of the buffer content at a node is determined by the real-time traffic arrival process, the data traffic from upstream nodes, and the ability of the downstream nodes to absorb the data traffic. The predictive policy attempts to predict the buffer content of a node one round trip delay into the future. Based on this prediction, in each time period, a node informs its immediate upstream neighbors, of the maximum rate at which they can transmit data traffic. This rate must be low enough to keep losses low and high enough to keep the throughput high. We characterize the real-time traffic by its mean rate, its variance and the correlation. We study the performance of our scheme over a wide range of real-time traffic parameter values. We investigate the effect of higher link speeds, and larger propagation delays on the performance of the network. We study the unfairness that results when sources are at different distances from a common access node, and propose a solution to overcome the bias towards the near source. We also compare the predictive policy to other end-to-end control policies. Our investigation shows that the predictive control policy performs significantly better than the static or adaptive end-to-end rate control policies.