Current network slicing solutions suffer from poor inter-slice isolation, as the performance of one slice can be influenced by the traffic in other slices. New technologies such as Flex Ethernet can offer hard isolation via dedicated resources at the physical and MAC layers. However, to create cost-efficient hard slices in large 5G access networks, a "routing and slot allocation" must be solved quickly. While the underlying network design problem is not new, two extra constraints need to be considered: a specific order in slot activations and a bandwidth allocation policy with statistical multiplexing. We propose a compact and extended formulation to derive FlexE-CG, an algorithm based on columngeneration to solve large instances. We reinforce the extended formulation to improve the lower bound by deriving valid inequalities, and we provide necessary and sufficient conditions under which the inequalities are facet-defining. We show that these inequalities improve the lower bound by more than 20% on various IP-Radio Access Networks (RAN). We also show that FlexE-CG can provide solutions within an optimality gap of 10% in a few minutes.