We study information theoretic models of interference networks that consist of K Base Station (BS) - Mobile Terminal (MT) pairs. Each BS is connected to the MT carrying the same index as well as L following MTs. We fix the value of L and study the per user Degrees of Freedom (puDoF) in large networks. We assume that each MT can be associated with N BSs, and these associations are determined by a cloud-based controller that has a global view of the network. An MT has to be associated with a BS, for the BS to transmit its message in the downlink, or have its decoded message in the uplink. We propose puDoF inner bounds for arbitrary values of L when only the uplink is considered, and characterize the uplink puDoF value when only zero-forcing schemes are allowed. We then introduce new achievable average uplink-downlink puDoF values, and show their optimality for the range when N \leq (L/2) and when we restrict our attention to zero-forcing schemes. Additionally, for the remaining range, we characterize the optimal downlink scheme when the uplink-optimal associations are used. Finally, we show that the proposed scheme is information theoretically optimal for Wyner's linear interference network.