This paper considers a downlink (DL) system where non-orthogonal multiple access (NOMA) beamforming and dynamic user pairing are jointly optimized to maximize the minimum throughput of all DL users. The resulting problem belongs to a class of mixed-integer non-convex optimization. To solve the problem, we first relax the binary variables to continuous ones, and then devise an iterative algorithm based on the inner approximation method which provides at least a local optimal solution. Numerical results verify that the proposed algorithm outperforms other ones, such as conventional beamforming, NOMA with random-pairing and heuristic-search strategies.