In this paper, a fault tolerance theory for queue based congestion control is proposed in the behavioral system theoretical framework. The router dynamics changes instantaneously when the fault occurs causing the undesirable instability in the queue length. To improve the dynamics of TCP/AQM networks, suitable controllers (stable and optimal) are synthesized using the concept of canonical controller in real time. This makes the network to maintain its stability. The main feature of this fault accommodation technique is that it solely relies on the input-output trajectories generated by the actual plant model (without having its exact knowledge) and on the performance specifications given by the desired behavior. The least restrictive controller satisfying the desired behavior is chosen based upon the generated trajectories.