Research On Scheduling Disciplines With Self-Similar Traffic Input

With the fast increment of the network transmition service types, Internet is required to offer different quality of services (QoS) for different traffic. The packet scheduling discipline is one of the key mechanisms to realize QoS control of the network traffic and is the important part of network resource management. The traffic with different QoS requirement can get different services by using packet scheduling disciplines which control the use of link bandwidth among different packets. On the other hand, a series of recent measurement analysis results indicate that network traffic exhibit ubiquitous properties of self-similarity and long-range dependence(LRD). The traditional network models neglecte this important characteristic while describing the real network traffic, resulting in that the characteristic of actual network traffic can't be portrayed by traditional network model. Packet scheduling discipline and self-similar (LRD) nature of network traffic have become two hot questions both at home and abroad.In this thesis, the packet scheduling disciplines with self-similar traffic input are studied. The mojor works are as follows:Firstly, on the basis of understanding the basic principle of queueing theory and traditional traffic model, the traditional short-range dependence traffic and modern self-similar traffic are compared and studied. Self-similar characteristic of network traffics is depicted,and several mathematical definitions of self-similarity are given. Influences of self-similar characteristic on scheduling disciplines are summarized.Secondly, the available packet scheduling disciplines in wire networks are deeply studied. On the basis of analysis and summarization of their principle, function, classification and performance indix of packet scheduling disciplines, static priority scheduling,round robin scheduling and fair queueing scheduling based on GPS model are researched in detail. Theoretical performance assertment of the fair queueing scheduling algorithms based on GPS model are given.The key content of the thesis is constructing the simulation model of scheduling disciplines with self-similar traffic input on OPNET platform. Impact of self-similarity on scheduling disciplines is researched. Simulation models of four types scheduling disciplines, namely PQ, RR, WRR and WFQ are builded. Based on the OPNET simulation driven by the traces generated with self-similar traffic model ,the performances of different scheduling disciplines with short-range dependence traffic and self-similar traffic input are compared and analyzed .The author focuses on the study of performance influence factors of scheduling disciplines (namely delay, loss rate) under different conductions of self-similarity (self-similar parameter H) and service characteristics (buffer space and utilization ratio) , and get some useful conclusions which can be used to furnish some useful tools for the design and analysis of scheduling disciplines.Finally, the problem of WRR scheduling discipline with self-similar traffic input is analyzed, and a improved WRR scheduling discipline is proposted .Simulation results show that with self-similar traffic input, the performances of delay and loss rate of the improved scheduling discipline are superior to the original scheduling discipline.