Parallel simulation techniques for telecommunication network modelling

Abstract

In this thesis, we consider the application of parallel simulation to the performance modelling of telecommunication networks. A largely automated approach was first explored using a parallelizing compiler to speed up the simulation of simple models of circuit-switched networks. This yielded reasonable results for relatively little effort compared with other approaches. However, more complex simulation models of packet- and cell-based telecommunication networks, requiring the use of discrete event techniques, need an alternative approach. A critical review of parallel discrete event simulation indicated that a distributed model components approach using conservative or optimistic synchronization would be worth exploring. Experiments were therefore conducted using simulation models of queuing networks and Asynchronous Transfer Mode (ATM) networks to explore the potential speed-up possible using this approach. Specifically, it is shown that these techniques can be used successfully to speed-up the execution of useful telecommunication network simulations. A detailed investigation has demonstrated that conservative synchronization performs very well for applications with good look ahead properties and sufficient message traffic density and, given such properties, will significantly outperform optimistic synchronization. Optimistic synchronization, however, gives reasonable speed-up for models with a wider range of such properties and can be optimized for speed-up and memory usage at run time. Thus, it is confirmed as being more generally applicable particularly as model development is somewhat easier than for conservative synchronization. This has to be balanced against the more difficult task of developing and debugging an optimistic synchronization kernel and the application models.