Router Performance Analysis Based on a Periodical Logarithmic Traffic Model
Xie Gao · Chinese Journal of Computers · 2002
Traffic modeling is fundamental for network design and performance analysis. Various traffic models have been proposed, but their applicability is challenging. A periodical logarithmic traffic model is proposed based on two collects of data from network test and measurement in this paper. One experimental traffic data is from the SUN Web server by analysis system log file,which has been widely used by many international authors, and another is collected from SNMP MIB(Management Information Base) polling on routers and packet capturing in the ICT(Institute of Computing Technology) network. Based on the experimental data, the traffic rate is supposed to be bounded by R(t)=αln(βt+1), where α and β are constant. Generally α indicates magnitude of traffic rate, while β shows some kind of traffic increasing. Obviously, the traffic model is very different from traditional statistical ones. Since the maximum traffic rate is considered, there is no statistical supposition assumed to the traffic model. Periodicity and burst of traffic are emphasized. The constants, α and β, are found from the time characterization of bound of burst by function fit on traffic profile by measurement on operating network.Designing issues for routers,as the basic unit for Internet, such as ability of transmission,input and output interfaces, buffer size and ability of processing of router fabric are discussed based on the traffic model.