Recent technology has led to systems with architectures that distribute the controller effort over sub-controllers to respect information flow and/or resource constraints. The associated communication uncertainty between sub-controllers partly governs the performance of the controller. The related controller synthesis methodology has to address internal stability concerns and has to incorporate the effect of communication uncertainty into the performance metric. In this article, it is demonstrated that different canonical distributed architectures derived from a centralized design can result in appreciably different performance. Architectures of information flow that result in convex optimization problems are developed and synthesis methods to incorporate robustness with respect to model uncertainty of communication channels are obtained. The efficacy of the approach is demonstrated on application problems.