Edge-to-Edge Aggregate Flow Control for the Internet

We present an edge-to-edge flow control scheme which enables an ISP to achieve weighted max-min fair bandwidth allocation among all source-destination pairs on a per-aggregate basis within its network. The motivation behind the scheme is the absence of per-aggregate flow control in the current Internet, resulting in inability to enforce a certain fairness on source-destination flows. The proposed flow control scheme is hierarchical in that in the upper layer weighted max-min flow control is implemented and acting on a per-aggregate and edge-to-edge basis and in the lower layer TCP flows belonging to a source-destination flow share its per-aggregate bandwidth allocated by the upper layer in their normal way. Thus, the scheme requires neither modification nor replacement of TCP congestion control. The distributed algorithm to compute weighted max-min fair rates is based on PI control. It is scalable in that the computational complexity imposed on each link is O(1), i.e., independent of number of aggregate flows travelling through it, stable in that it converges asymptotically to the desired equilibrium satisfying the minimum plus weighted max-min fairness, and has explicit link buffer control in that the buffer occupancy of every bottlenecked link in the network asymptotically converges to the pre-defined value. By appealing to the Nyquist stability criterion, we mathematically prove the asymptotic stability of the algorithm in presence of aggregate flows with different round-trip delays. Through extensive simulations we demonstrate the effectiveness of the proposed scheme in controlling per-aggregate flows.