Adaptive REDn gateway and adaptive token bucket traffic conditioner for an assured rate service for TCP

• Main Author: Liu, Philip K. M.
• Language: English
• Published: 2009
• Online Access: http://hdl.handle.net/2429/11386

The majority of traffic in the Internet today is best-effort traffic. Best-effort traffic is believed to continue predominating over traffic of the emerging enhanced network services that needs various QoS guarantees. The key congestion controls developed for best-effort traffic forwarding in the Internet are TCP and RED. Although TCP is simple and time-invariant, the received throughputs of TCP flows vary with the associated RTTs, as opposed to the throughput expectations, to result in a bias against the flows with relatively long RTTs and large maximum windows. In general, the nature of the TCP flows traversing a gateway, such as the number of active flows and the sizes of the congestion windows, varies over time, and the per-flow throughput expectations span a wide range. However, RED merely makes a fairly limited linear adaptation to the varying nature, and virtually neglects the throughput expectations of the flows. This thesis introduces Adaptive REDn Gateway and Adaptive Token Bucket Traffic Conditioner for rectifying these shortcomings of TCP and RED: The architecture of Adaptive REDn Gateway and Adaptive Token Bucket Traffic Conditioner, with throughput-based deterministic traffic conditioning performed at the boundary and delayed-policing applied in the interior of the network, enables the implementation of an assured rate network service for TCP in the Internet, providing a short queuing delay and an assured bandwidth distribution at each gateway. The overall performance of the mechanisms is evaluated with numerous simulations in a wide variety of scenarios, and all the results certainly corroborate that the architecture attains the goals.