In this paper we propose a randomized primal-dual proximal block coordinate
updating framework for a general multi-block convex optimization model with
coupled objective function and linear constraints. Assuming mere convexity, we
establish its $O(1/t)$ convergence rate in terms of the