Doctor of Philosophy
Lawrence, Jeremy Robert, Performance analysis of intelligent networks, Doctor of Philosophy thesis, , University of Wollongong, 1994. http://ro.uow.edu.au/theses/2103
The Intelligent Network architecture is an extension of the existing telephone network architecture to provide centralised definition and control of services. This architecture allows enormous flexibility in introduction of new services, and allows customers to define and control their own services with a large degree of independence from network providers.
This thesis is concerned with systematic modelling and analysis Intelligent Networks and Intelligent Network services.
There is a significant body of prior art on queueing network modelling of the signalling networks and switching equipment which comprise Intelligent Networks. The first part of the dissertation defines a scheme for systematically translating intelligent network service definitions into parameters for such queueing network models. A representation syntax for services is defined for this purpose, and the steps required to translate a service to analysable form are defined with the aid of this syntax.
The second part of the dissertation extends existing queueing network analysis techniques to obtain certain performance measures which are important in the analysis of Intelligent Networks, and which are lacking in previous work. In particular, the calculation of completion time distributions for entire call set-ups, as opposed delays at individual devices, is calculated.
Telecommunications switches may use simple priority-based approaches to preserve throughput under congestion conditions. Such measures have previously been observed to cause oscillatory behaviour in voice switches, through a mechanism known as "traffic synchronisation". Similar behaviour is observed in a seemingly straight-forward model of a signalling network supporting Intelligent Network transactions. This is investigated fully in the third part of this thesis. Traffic synchronisation in Intelligent Networks leads to connection establishment delays significantly greater than those predicted by standard queueing analysis techniques. A n analysis technique quantifying these delays is developed and validated by comparison with simulations.
The trade-off between prevention of traffic synchronisation at normal loads and preserving throughput at overloads is also considered. A congestion avoidance technique which prevents traffic synchronisation at little performance cost is described and analysed.