Font Size: a A A

System Model And Flow Control Mechanism For Wireless Networks With Heavy-tailed Traffic

Posted on:2006-04-30Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y WangFull Text:PDF
GTID:1118360182983320Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
Although Internet access through UMTS networks is a special usage scenario, itpresents a general network convergence problem, in which wireless and wirednetworks are connected to provide wireless Internet service. It has been a hot topic forboth academic circle and industry. And the Internet heavy-tailed traffic makes it anextremely challenging task to analyze and evaluate the system performance.We present a novel flow control scheme with coarse granularity. In this scheme,a scheduling agent is deployed at the edge of wireless network. By exploiting thedelay-insensitive and rate-insensitive characteristics of the predominant Internetelastic traffic, the new scheme significantly decreases the average response time ofweb access, without sacrificing the utilization of wireless bandwidth. Taking intoconsideration that the UMTS systems prefer fixed rate within burst data transfer, thescheme uses files as the basic scheduling units, avoiding dynamical rate adjustmentduring file transmissions.In order to make the scheme independent to traffic models, the systemperformance under different traffic scenarios is systematically evaluated. Based onthe evaluation, a novel methodology is proposed to deal with Internet traffic. Themethodology decomposes the Internet heavy-tailed traffic into long flows and shortflows, with different system models, respectively. Based on the evaluation of therelationship between the traffic decomposition and the system performance, weconclude that the performance of the short flow system shapes the outline of theoverall system performance, while long flows can be regarded as disturbance. Basedon this conclusion, we deduce the formulas for the blocking probability and theaverage queuing delay under heavy-tailed traffic for multi-server systems withimpatient customers.Furthermore, by introducing diffusion approximation on short flows, we proposea novel analytical model for the heavy-loaded multi-server queues, in which the Itointegral is used to deduce the differential equations. This differential equationscharacter the relationship between the blocking process, the servers' idle process andthe queuing process. Based on this model, we study the dependency of the blockprobability on the variance of short flows, and on the number of servers. In addition,the model is a novel theoretical tool for dynamic performance analysis and controlalgorithms design.At last, based on the analytical tools and the theoretical results, a practicalcoarse-grained flow control algorithm is proposed. By using the feedback controlmechanism, the algorithm is stable under different traffic patterns and perturbationparameters. And by introducing feed-forward control mechanism, the system isinvulnerable to the burst of long flows. Simulation results show that the algorithmimproves the system performance under different traffic scenarios.
Keywords/Search Tags:Wireless Internet access, Coarse-grained flow control, Heavy-tailed traffic, Diffusion approximation, Ito Integral
PDF Full Text Request
Related items