Quantifying the impact of correlated failures on system reliability by a simulation approach

Correlation poses a serious threat to many engineered systems because the simultaneous failure of multiple components can dangerously degrade performance. Given the high cost of system failures in business and mission-critical applications, methods to explicitly consider the impact of correlation on system reliability are essential. This paper constructs a stochastic-flow network model to analyze the performance of a computer network, where there exists correlation between the failures of all the physical lines and routers comprising the edges and nodes of the network. That is, we address global-scale events that can cause widespread damage to the performance of the network. We propose a simulation approach to estimate the probability that a given amount of data can be sent from a source to sink through this network. This probability that the network satisfies a specified level of demand is referred to as the system reliability. Experimental results demonstrate that correlation can produce a substantial impact on system reliability. The proposed approach, thus, captures the influence of correlation on system reliability and offers a method to quantify the utility of reducing correlation.