Browse Theses

Self-similar characteristics have been discovered in the Ethernet LAN traffic, Variable-Bit-Rate (VBR) video traffic, MPEG4 traffic, and World Wide Web (WWW) data transfer. Unlike conventional traffic characteristics, such as the Poisson traffic model, the self-similar traffic model has the ability to capture traffic fluctuations over many time scales. This traffic has “bursty” traffic patterns that tend to exhibit certain degrees of correlation between arrivals, and show long-range dependence over time. Traffic burstiness can be identified from the traffic's Hurst (H) parameter. Traffic is categorized as self-similar traffic, if its Hurst parameter is between 0.5 and 1.

Traffic modelling and network performance have all been influenced by the discovery of self-similar traffic. The purpose of the thesis was to investigate the implications of self-similar traffic on ATM network performance and to design strategies to achieve better network performance for the self-similar traffic model. The thesis focused on the impact of the Hurst parameter on two parameters of Quality of Service (QoS), notably end-to-end cell delay and end-to-end cell loss. Effective bandwidth was performed and statistical multiplexing strategies were designed. The Random Midpoint Displacement (RMD) and Successive Random Addition (SRA) methods as a self-similar synthetic traffic generator were compared and evaluated.

Results indicated that the RMD method is more reliable in generating self-similar traffic than the SRA method. Statistical analysis used during this investigation, applied the G/D/1 queuing model with First In First Out (FIFO) discipline service. The simulation models adopted the Discrete Time Advance algorithm and were also performed in OPNET 7.0. Additionally, the inequalities for Mills' ratio were proposed to derive Cell Loss Probability and effective bandwidth lower bounds. In general, numerical and simulation results showed a good agreement, which demonstrated that the network performance had an adverse relationship with the Hurst parameter. This performance degradation however, was alleviated to some extent by implementing the traffic-shifting and shuffle methods as statistical multiplexing strategies for the self-similar traffic. The investigation might be developed by involving non self-similar traffic models in ATM network and in other network models such as in the Internet.