Research On Flow Control Technolngy In Supanet

This thesis is dedicated to flow control techniques in SUPANET, an NGI framework based on a Single-layer User-data switching Platform Architecture (SUPA) and developed at Sichuan Network Communication Key Laboratory. The basic idea behind SUPA is to embed all necessary functionalities in a simplified user-data switching platform to meet the challenges faced with existing Internet, i.e. high-speed switching, QoS-enabling, Security, mobility, and etc.Many techniques have been used in networks to deal with issues more or less related to congestion, such as congestion control, traffic engineering, and flow control, although they differ from each other. Congestion control focuses on preventing a subnetwork from being overwhelmed by excessive traffic and traffic engineering emphasizing on distribution of traffic load across links in the subnetwork. Flow control, on the other hand, concerns how receivers can control the transmission rate of the senders between two directly connected nodes or over an abstract end-to-end connection.The "Sliding Windows" is a typical flow control mechanism in early networks but unsuitable for high-speed networking environment or long distance communication since it requires a very large window size. Therefore, a "pause" mechanism is introduced in Ethernet to replace "sliding windows" mechanism. A receiver simply sends a pause frame to upstream node via a port to ask to stop sending data for an assigned period of time. In a matter of speaking, this mechanism is "selfish" as it does not consider the possible impact on the upstream node and merely care to ease possible its own congestion.In the SUPANET, QoS is one of main concern and there is a need for a "humanized" flow control mechanism. This thesis has tentatively proposed a flow control policy called PPR-BER (Polite Pause Request and Best Effort Response). In other words, a "Pause" frame sent to an upstream node implies a polite "pause request" and the sender received a pause request will try its best to reduce data transmission without causing its own congestion. The simulation result with PPR-BER policy has been encouraging and the data loss rate with PPR-BER is obviously reduced with an acceptable increase in transit delay and jitter. This suggests that in a high-speed network flow control be a sensitive matter, especially for a QoS enabled network and trade-offs should be made among multiple QoS parameters.The work presented in this thesis is preliminary but it does open up a new thinking in dealing with flow control issues in a high-speed and QoS-enabled network.