Recently distributed integral controllers relying on averaging and communication have been proposed as effective means for optimal frequency regulation in power systems, load balancing of network flows, and as natural extensions to static consensus controllers. Typically, only the questions of stability, disturbance rejection, and steady-state resource allocation are addressed in the literature, and the problems of transient performance and optimal communication network design remain open. In this paper we consider the optimal frequency regulation problem and propose a principled heuristic to identify the topology and gains of the distributed integral control layer. We employ an l1-regularized H2-optimal control framework as a means for striking a balance between network performance and communication requirements. The resulting optimal control problem is solved using the alternating direction method of multipliers algorithm. For the IEEE 39 New England benchmark problem, we demonstrate that the identified sparse and distributed integral controller can achieve reasonable performance relative to the optimal centralized controller. Interestingly, the identified control architecture is directed and correlates with the generator rotational inertia and cost coefficients.