Congestion Control in ATM Networks
ATM networks promise to provide large bandwidth and QoS guarantees but only with effective traffic management schemes such as CAC and congestion control. We developed a method of testing the robustness of flow control algorithms and studied network topologies in which two different congestion control schemes coexist. We considered both pure ATM and IP-over-ATM networks.

To evaluate performance, we used our available bit rate (ABR) simulator, which can overlay any set of virtual circuit (VC) routes on any topology. We implemented 5 flow control algorithms, identified 10 performance metrics, and studied key behaviors: source-rate oscillations and interference in rate control by nonbottlenecked switch nodes. We established criteria for the ability of an algorithm to perform well regardless of changes in topology, feedback delays, or traffic. We identified 11 network topologies in which a flow control algorithm must be tested: few enough to be studied within a reasonable time, but numerous enough to cover all potentially important and/ or interesting scenarios.

We then evaluated the interoperability of pairs of ABR flow control algorithms in diverse ATM networks. We concluded that heterogeneous configurations with two different ABR flow control algorithms operating simultaneously cause users to receive disparate bandwidths if (1) the algorithm at a bottlenecked switch is much more oscillatory than the algorithm at the nonbottlenecked one(s); and (2) at least one VC traverses only the more oscillatory algorithm, and at least one VC traverses both algorithms.