This paper presents a fast, randomized greedy algorithm to solve the Component Commonality Problem approximately. It is based on a geometric random walk, that starts from a given initial solution and accepts only better points during the walk. We use a new type of analysis, that is not based on conductance, but makes use of structural geometric properties of the problem, namely the smoothness of the set of feasible points.