High-bandwidth traffic aggregates may occur during times of flooding-based distributed denial-of-service attacks or flash crowds. Congestion control of these traffic aggregates is important to avoid congestion collapse of network services. This paper presents a class of feedback-control algorithms that proactively protect a network server from overload by installing rate throttles in a set of upstream routers. A control-theoretical framework is proposed to optimize the control setting such that throttling can be achieved in a distributed and fair manner. We develop control-theoretic algorithms that (1) are highly adaptive by avoiding the configuration of unnecessary control parameters, (2) provide max-min fairness for any number of throttling routers, (3) respond very quickly to network changes, (4) are extremely robust against extrinsic factors beyond the system control, and (5) are stable under given delay bounds.