We present a framework for hierarchical traffic grooming in mesh networks with the objective of minimizing the total number of electronic ports. At the first level of hierarchy, we decompose the network into clusters and designate one node in each cluster as the hub for grooming traffic. At the second level, the hubs form another cluster for grooming inter-cluster traffic. We view each (first- or second-level) cluster as a virtual star, and we present an efficient near-optimal algorithm for determining the logical topology of lightpaths to carry the traffic within each cluster. Routing and wavelength assignment is then performed directly on the underlying physical topology. Our approach scales to large network sizes, and facilitates the control and management of multigranular networks. Comparisons to lower bounds indicate that it is also efficient in its use of the network resources of interest, namely, electronic ports and wavelengths.