We investigate the optimal partitioning of the end-to-end network QoS budget to quantify the advantage of having a non-uniform allocation of the budget over the links in a path. We formulate an optimization problem that provides a unified framework to study QoS budget allocation. We examine the underlying mathematical structure for the optimal partitioning and dimensioning equations. In the context of network dimensioning, we then show that optimal partitioning can bring large cost reductions as compared with equal partitioning based on the results on small networks.More importantly, we also find that optimal partitioning gives significant improvements in robustness in the presence of failed components and in fairness when the traffic demand is different from the forecast, two effects that had not been observed in previous work and that can have a significant effect on network operations.