Accelerated simulation of ATM switching fabrics.

Abstract

Asynchronous transfer mode (ATM) is accepted as the solution to the broadband integrated services network known as B-ISDN. The information transmitted through an ATM network is divided into fixed sized packets, called cells. These cells traverse an ATM switching fabric, wherein they are electronically switched a number of times in order to arrive at the correct destination. For design purposes, it is important to know the cell loss ratio (CLR) for the buffers in an ATM switching fabric. We model the ATM switch, as an intree acyclical network of queues receiving N independent heterogenous Markov modulated input streams of cells. Due to the complexity of the model, simulations must be used to find the CLR for a hot spot buffer in the switching fabric. Crude Monte-Carlo simulations are not very efficient, since most services require a very small CLR, e.g. 10$\sp{-9}.$ Importance sampling techniques must be used. To make cell losses less rare, we increase (exponentially twist) the arrival rate of cells directed toward the hot spot buffer. We use the twisted probabilities with a splitting importance sampling technique to obtain an efficient estimate of the CLR for this buffer.