The paper proposes an optimization approach to flow control, aiming to maximize the aggregate source utility over transmission rates. The authors view network links and sources as processors in a distributed computation system, solving the dual problem using a gradient projection algorithm. Sources select transmission rates that maximize their benefits (utility minus bandwidth cost), while network links adjust bandwidth prices to coordinate the sources' decisions. The algorithm allows for different, substantial, and time-varying feedback delays, and updates at different times and frequencies. Asynchronous distributed algorithms are provided, and their convergence is proven in a static environment. Experimental measurements from a prototype illustrate the algorithm's convergence in a slowly time-varying environment. The paper discusses the fairness property of the algorithm and its extension to a multicasting environment.The paper proposes an optimization approach to flow control, aiming to maximize the aggregate source utility over transmission rates. The authors view network links and sources as processors in a distributed computation system, solving the dual problem using a gradient projection algorithm. Sources select transmission rates that maximize their benefits (utility minus bandwidth cost), while network links adjust bandwidth prices to coordinate the sources' decisions. The algorithm allows for different, substantial, and time-varying feedback delays, and updates at different times and frequencies. Asynchronous distributed algorithms are provided, and their convergence is proven in a static environment. Experimental measurements from a prototype illustrate the algorithm's convergence in a slowly time-varying environment. The paper discusses the fairness property of the algorithm and its extension to a multicasting environment.