Congestion control ensures that network resources are divided fairly and efficiently among competing connections. Although congestion control has been studied by researchers for a long time, very high-speed networks using ATM pose a new set of challenges.

The proposed credit-based mechanism provides flow control tailored to ATM networks. Simulation, analysis and experiments on switching hardware have shown that for a wide variety of traffic patterns, credit control is fair, uses links efficiently, minimizes delay, and guarantees no cell loss due to congestion. The credit system is especially well suited to data traffic that is bursty, unpredictable, and has little tolerance for delay.

Other approaches to flow control, including rate-based flow control, may require less expensive hardware and may be effective for steady traffic, but do not handle bursty traffic well. While no one can predict what kind of traffic will dominate future networks, recent evidence suggests that it will be bursty. Thus a major challenge for network research will be to find congestion control mechanisms that blend the hardware simplicity of rate-based flow control with credit-based flow control's ability to handle bursts. This research, in turn, will depend on more experience with real applications and high-level protocols running over ATM.

The high-speed networking market is solidifying quickly, and basic issues in congestion control should be well understood before the market is forced to choose a solution with unknown properties. The lessons learned from the credit-based approach should be incorporated into what- ever flow control system that is finally standardized or widely implemented.

PostScript, PDF.