Network services in 5G will be rolled out as pools of virtual network functions (VNFs) exploiting the advantages of both software-defined networking and network function virtualization. In this context, 5G network services are envisaged as ordered sequences of VNFs resulting in the so-called VNF Forwarding Graphs (VNFFGs). Such VNFs can be allocated over a number of distributed but interconnected data centers (DCs). In this work, a cloud/network orchestrator is discussed to dynamically process and accommodate VNFFG requests over a pool of DCs interconnected by a multi-layer (packet/flexi-grid optical) transport network infrastructure. Two different cloud and network resource allocation algorithms are proposed aiming at: (1) minimizing the distance between the selected DCs and (2) minimizing the load (i.e., consumed cloud resources) of the chosen DCs. In the performance evaluation, the proposed algorithms are experimentally validated and compared on the CTTC ADRENALINE testbed.