A comparative study of flow control methods in high-speed networks

The authors apply fluid modeling techniques to the analysis of data flow on high-speed networks. The network is modeled as a single-hop high-speed virtual channel (VC). The VC shares server assets with cross-traffic, thereby diminishing the server assets available to the VC. Transmission line delay is inserted in the feedback loop to control the VC source rate. Cross-traffic is modeled as a twice reflected Brownian motion random process. Three previously proposed control schemes are compared which are rectangular sliding window, linear increase/exponential decrease, and throughput maintaining. Next, a new control scheme, the slope controller, and a variation of it, the weighted average controller, are also considered. Simulation of the VC was performed with each control scheme and qualitative comparisons are made between the five control schemes. The results indicate that the weighted average admission controller provides the minimum end-to-end delay through the virtual channel.<
>