In Part 1 of this two-part paper, we developed an adaptive neural predictive control mechanism for resource management at a network node. Neural predictive models which account for the control-loop delays in the information transfer process were developed and analytical techniques for the control-signal computations using gradient functions of the neural models were given. In this second part, we describe closed-loop control procedures for sources transmitting data to a single bottleneck network node with periodic exchange of control information. A control algorithm runs at the bottleneck node with the task of distributing the available bandwidth between the data sources. The goal is to control at any time instant the rates of the incoming data flows so as to maximize network utilization and at the same time prevent or react to network congestion.