A recursive variance-reduction algorithm for estimating communication-network reliability

In evaluating the capacity of a communication network architecture to resist possible faults of some of its components, several reliability metrics are used. This paper considers the 𝒦-terminal unreliability measure. The exact evaluation of this parameter is, in general, very costly since it is in the NP-hard family. An alternative to exact evaluation is to estimate it using Monte Carlo simulation. For highly reliable networks, the crude Monte Carlo technique is prohibitively expensive; thus variance reduction techniques must be used. We propose a recursive variance-reduction Monte-Carlo scheme (RVR-MC) specifically designed for this problem, RVR-MC is recursive, changing the original problem into the unreliability evaluation problem for smaller networks. When all resulting systems are either up or down independently of components state, the process terminates. Simulation results are given for a well-known test topology. The speedups obtained by RVR-MC with respect to crude Monte Carlo are calculated for various values of component unreliability. These results are compared to previously published results for five other methods (bounds, sequential construction, dagger sampling, failure sets, and merge process) showing the value of RVR-MC