Aggregate Rate Control for Efficient and Practical Congestion Management

Active queue management (AQM) promises to overcome the current limitations of end-host only congestion control by providing congestion feedback information before router queue buffers overflow. Many emerging AQM approaches use proportional integral (PI) controller design because of PI’s simplicity and effectiveness. Unfortunately, these promising AQMs still face a critical deployment challenge since there are no simple and effective PI control parameter configurations available for time-delayed systems (i.e. the Internet). As a solution, we present the Aggregate Rate Controller (ARC), a reduced parameter PI controller for Internet traffic. ARC, founded on both classical control theory and a sound understanding of Internet congestion control, uses a low frequency rate-based approach to detect congestion that minimizes control noises and provides more flexible link Quality of Service (QoS) compared with queue-based approaches. In addition, we provide practical configuration guidelines for ARC that produce efficient and resilient performance over a wide range of traffic conditions. Simulations verify that ARC effectively handles network congestion over a range of network and traffic conditions, overall outperforming other mechanisms in terms of queue dynamics, link utilization, data loss rate and object response time for Web traffic.