Massive MIMO technique can effectively increase system capacity in the fifth Generation (5G) cellular network, where Base Station (BS) is equipped with a very large number of antennas. Considering the impact of channel estimation error on performance, the transmission power minimization problem is formulated subject to the non-outage probability constraints of each users signal to interference plus noise ratio. In respect that the non-convex probability constraints make the downlink beamforming difficult to solve, Uplink-Downlink Duality Algorithm (UDDA) is proposed to design Coordinated BeamForming (CBF) by using the property of trace of the matrix to scale the non-convex probability constraint. To reduce the signaling overhead in Massive MIMO system, a Distributed Algorithm based on Large System Analysis (DALSA) is proposed, which only needs the large-scale channel information. The simulation results show that DALSA, in the targeted SINR constraint, not only reduces instantaneous channel state information transmission overhead in Massive MIMO system, but also performs well in robustness compared with UDDA.