In a cellular wireless network a choice is often available of which base station serves a user on each channel, where the best selection can significantly improve network efficiency. In this work we present a downlink distributed network scheduling (DNS) algorithm that specifies a limited information exchange between base stations to improve scheduling decisions. The algorithm is provably optimal in the sense that it converges to the network throughput proportional fair solution. The implication of more general forms of user utility is also discussed. We present both an idealized and a practical version of the DNS algorithm. In the former we assume that all users can be served by all base stations (but not simultaneously) on each channel, so that optimal allocation can be achieved entirely through coordination of base station scheduling priority. In the latter the decision-making is shared between users and base stations, where users perform server selection (NLB) on each channel, the channels are allocated by a channel management (CM) algorithm, while the base stations coordinate scheduling priority across the resulting data paths. We show the sense in which the practical version of DNS achieves the optimal allocation, and briefly mention a commercial system where these algorithms are being proven in the field. An interference-mitigation technique is introduced, referred to as adaptive reuse (AR), in which the transmit power of base stations is adjusted separately on each channel to create reuse patterns appropriate to the network topology and load. The resulting channel quality heterogeneity can be exploited by DNS to improve overall network efficiency and allocation flexibility. But well-designed AR patterns must account for receiver interference nulling at the users, which DNS also exploits, and the resulting tradeoffs are non-obvious.