This paper contributes a customer-stationary probabilistic model that can be used to measure the availability of a variety of replication control methods. This includes methods that tolerate site failures and those that tolerate both site failures and network partitioning. Existing availability measures assume a k-out-of-n up/down failure modeling, and are therefore not useful in evaluating methods that handle network partitioning. Our model relies on an s π probability failure modeling, which enumerates the probability that a site s belongs to a particular network partition π. We demonstrate our model through a study of the effect of the degree of replication on the read-one-write-all and the quorum consensus methods. We also analyze the model's computational complexity when used in fully connected and ring network topologies.