In current multi-domain networks relying on the path computation element (PCE) architecture, domain sequence computation can be performed through mechanisms which may fail to guarantee both efficient resource utilization and an adequate level of confidentiality and scalability. In this study, we first propose a hierarchical instance of a path-vector protocol, called the domain sequence protocol (DSP), dedicated to provide the PCE with effective domain sequence information. Then we consider and evaluate through simulations the performance of the integrated DSP-PCE architecture. Results show that, compared to current routing solutions based on the border gateway protocol, significant improvements can be achieved in terms of the overall network resource utilization. In addition, this study identifies the most suitable DSP-PCE features for different network scenarios, including the announcement of single or multiple routes per domain, the advertisement of aggregate or maximum reservable inter-domain bandwidth information, the use of different PCE-based path computation procedures, and the possibility to perform additional computation attempts along different domain sequences. Finally, the experimental implementation of the proposed DSP-PCE architecture is provided to show its feasibility and its fast convergence time in the range of a few milliseconds.