In this paper, we propose a dynamic optimization approach to end-to-end flow control in data networks. The objective is to maximize the aggregate utilities of the data sources over soft transmission rate bounds and delay constraints. The network links and data sources are considered as processors of a distributed computational system that has a global objective function. The presented model works with different shapes of utility curves under the proposition of elastic data traffic. The approach relies on real-time observations of the delay as a measure of the data network congestion at the routers (network nodes). A primal-dual algorithm carried out by the data sources is used to solve the optimization problem in a decentralized manner. The calculated transmission rates are bounded and the sources are subjected to a maximum number of data packets that can be queued downstream of each transmission session. The algorithm solves for the rates without the access to any network global information while each source calculates its transmission rate that should maximize the global objective function. The calculated optimal rates conform to rate-to-queue proportionality. Finally, we present an extensive simulation results to demonstrate the reliability of the algorithm.