A. Kiss, J. Levendovszky, L. Jereb, ``Stratified Sampling Based Network Reliability Analysis'', in Proceedings of the 8th International Conference on Telecommunications Systems, Nashville, TN, USA, March, 2000, pp. 367-375 [psfile]

This paper treats resilience techniques and calculate numerical reliability measures by the means of stratified sampling for communication networks. In stratified sampling, the state space over which the estimation is carried out is divided into classes and a sample size is dedicated to each class. Estimation is then performed by calculating conditional statistical parameters in these classes. The efficiency of this method, however, hinges upon the sample sizes allocated to different classes. As will be seen, the main bottleneck of the applicability of stratified sampling stems form the unknown variances, which are needed for calculating the optimal sample allocation scheme. To overcome these difficulties the following methods will be proposed and numerically analysed:

· stratified sampling with sample allocation proportional to the class probabilities;

· "pre-estimation/post-processing scheme" in which the variances are estimated on a pre-defined sample size in advance (which helps to set the sample allocation scheme) and then the conditional expected values are estimated, based on the obtained sample allocation scheme;

· adaptive stratified sampling, which estimates the variances batch by batch, therefore the sample allocation scheme is updated after every batch estimation.

The performance of the methods and their numerical complexity will be thoroughly discussed and tested by extensive simulations.

Keywords: Network reliability and availability estimation, stratified sampling, Monte-Carlo methods.