The redundancy allocation problem is the problem of finding an optimal allocation of redundant components subject to a set of resource constraints. The problem studied in this paper refers to a series-parallel system configuration and allows for component mixing. We propose a new modeling/solution approach, in which the problem is transformed into a multiple choice knapsack problem and solved to optimality via a branch and cut algorithm. The algorithm is tested on well-known sets of benchmark instances. All instances have been solved to optimality in milliseconds or very few seconds on a normal workstation.