Self-similar traffic engineering and applications to mobile radio networks
Thesis DisciplineElectrical Engineering
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
Mobile networks are experiencing exponential rates of subscriber growth worldwide. In addition they are rapidly developing sophistication and capabilities for delivering multiple service types at widely varying data rates. Despite this very little is known about the traffic characteristics of such networks or the effects of such characteristics on network performance. There is a growing body of evidence that video and data traffic is not Markovian and cannot be modelled using the conventional models developed for wireline telephone networks. Instead these kinds of traffic are characterised by stochastic self-similarity and are highly bursty in nature. This bursty nature is known to significantly degrade network performance, but few analytic results quantifying this are available. Assuming that the self-similar nature of packet oriented voice and data traffic will hold when it is carried on mobile networks, this thesis develops the analysis of fundamental network elements, basic queues and random access channels with packet arrivals using a simple mathematical model of a self-similar process. These results quantify the degradation in network performance through increasing blocking and delay caused by the shift from random Poisson arrivals to self-similar arrivals and the further effects as the self-similarity parameter increases. Also presented here is the analysis of data sets taken from live New Zealand based cellular telephone networks. This analysis is limited by the data sets available but shows that where hand over traffic is high the channel seizure process appears not to be Markovian and may be self-similar. The thesis concludes with a discussion of the difficulties in obtaining valid results from computer simulation when the output data stream is self-similar. This is caused by the unique statistical nature of such sequences.