Global Optimality in Distributed Low-rank Matrix Factorization