Speaker: Yu-Fu Huang Advisor: Dr. Kai-Wei Ke Date : 2014, Mar. 17 A page-oriented WWW traffic model for wireless system simulations
Outline Traffic models from Poisson to Self-Similar WWW Traffic structure Web traffic characterization Simulation and results Conclusion Reference
The interest towards traffic model Traffic models are needed as input in network simulation. A good traffic model may lead to a better understanding of the characteristics of the network traffic itself.
Stochastic Counting Process Poisson process ⊆ Renewal process Independent increment Memoryless property Inter-arrival time pdf: Exponential Renewal process Independent increment Inter-arrival time pdf: Arbitrary X1X2X3 X4 X5X6 X7 T=X1+X2+X3+X4+… X1,X2,X3… are i.i.d Poisson process: - Any point in the time axis meets Memoryless property. Renewal process: - Only point exactly at exiting one period and entering a new period meets Memoryless property. t
Variance of sample mean approaches to zero as n approaches to infinite.
Traffic models from Poisson to Self-Similar Self-Similar process Long Range Dependency Infinite Variance
Heavy-tailed probability distribution
Outline Traffic models from Poisson to Self-Similar WWW Traffic structure Web traffic characterization Simulation and results Conclusion Reference
WWW Traffic structure Two approaches to data traffic modelling: Behaviorist or black-box approach: Modelled w/o taking into account the causes that lead to them Structure approach: Model design is based on the internal structure of traffic generating system
Outline Traffic models from Poisson to Self-Similar WWW Traffic structure Web traffic characterization Simulation and results Conclusion Reference
Pages per session
Time between pages
Page size
Heavy-tailed probability distribution
Packet size
Packet inter-arrival time Page Packet PIT
Outline Traffic models from Poisson to Self-Similar WWW Traffic structure Web traffic characterization Simulation and results Conclusion Reference
Test conditions 4MB Queue 2000s of average session interarrival time Constant service rate of 2 KBps 82.75s of average session interarrival time (I) (II) Test condition (I): With proposed model adapted to corporate environment Server utilization rate: 68% With ETSI model adapted to corporate environment Server utilization rate: 3% Test condition (II): With proposed model adapted to corporate environment Server utilization rate: 68% With ETSI model adapted to corporate environment but increasing average session interarrival time from 2000s to 82.75s Get server utilization rate: 68% Adjusted Utilization ESTI model
ESTI Model
Test condition (I)Test condition (II)
Conclusions (I) Traffic models summary: Independent interarrival time: Exponential Session or packet interarrival Cumulative independent interarrival time: Gamma or Erlang distribution Page interarrival Data size: Self-similar distribution Page size
Conclusions (II) ESTI model underestimates packet losses and delay in a queue due to the low load offered by the ESTI model. The proposed model generates a traffic load similar to the measured one and much more burstiness than the ESTI one.
Reference [2] Staehle D., Leibnitz K., and Tran-Gia P., “Source Traffic Modeling of Wireless Application” Institut für Informatik, Würzburg Universität, Technical Report No. 261, June [1] Reyes-Lecuona A., González-Parada E., and Díaz-Estrella A., “A page-oriented WWW traffic model for wireless system simulations” Proceedings of the 16th International Teletraffic Congress (ITC16), Edinburgh, United Kindom, pp , June [3] Michela Becchi, “From Poisson Process to Self-Similarity: a Survey of Network Traffic Models”