Relationships among delay, throughput, and fairness, three conflicting performance criteria in modern flow-controlled communication networks, are investigated, and an optimal tradeoff among them is defined. Several fair bandwidth sharing schemes are introduced and compared with each other for a variety of network configurations. The results obtained are then utilized to define an optimal operating point (OOP) which maximizes the generalized network power. Given a fixed routing, it is shown that the OOP can be found via any unconstrained optimization algorithm. If rerouting is allowed, it is also shown that the OOP can be found by solving a variant of the multicommodity convex cost network flow problem. An algorithm for this problem is developed and tested for a number of networks. A delay-directive flow control scheme is suggested to realize the OOP in real operating networks, and additional implementation issues are discussed.>