Traffic characterisation and performance optimisation of mobile networks.

Several recent studies show that network traffic is self similar, or exhibits long range dependent characteristics. Self similar traffic is problematic for routing and congestion control algorithms because self similar traffic is very different from conventionally considered traffic such as Poisson or Markovian traffic. Self similar behaviour is expected to occur in future data networks as well as can be seen in present networks carrying bursty services. Therefore self similar behaviour must be thoroughly understood if appropriate call admission controls, scheduling algorithms and congestion control mechanisms are to be designed. Also characteristics of data traffic play a crucial role in performance analysis and design of communication networks. Understanding the models of network traffic helps designing better protocols, better network topologies, better routing and switching hardware and provide better services to the users. Therefore the need of traffic characterisation is a major challenge faced by network engineers at present. This research illustrates the different service modelling distributions significantly changes the medium access control performance. This is validated against two popular service models for WWW browsing and Email connections. WWW browsing is modelled using heavy tailed Pareto distributed burst sizes characterizing self-similarity at the aggregate traffic level. Email sessions are presented with the Cauchy distributed connection sizes. The results conclude the different service modelling distributions have a significant impact on medium access control performances. Investigation of call admission control and scheduling algorithms for diverse service classes is also studied in this work. A novel admission control and scheduling algorithm-based on evolutionary algorithms is proposed and the superior performance of the proposed technique over the state of the art mechanisms is demonstrated on an example GPRS system