Service-oriented computing is considered today a key technology for the development of robust and high quality intelligent distributed and embedded applications. Extensive research and development in the past few years has pushed SOA technology into state-of-the-art application areas such as context-aware, cloud-connected, mobile enterprise systems. However, many of the critical components on building reliable, robust, and user-centric sensor-based SOA systems are still open for research. Hence, it is timely to reexamine SOA research opportunities and identify new research challenges for next generation SOA. One of the future SOA applications is large-scale cyber-physical systems (CPSs) that include deep interactions between cyber-sides and physical-sides through massive sensors, actuators, mobile devices, and computing servers connected by heterogeneous networks. The services of large-scale CPSs have challenging requirements such as accurate timeliness, high reliability, and dynamic adaptability. Many of the service components are deployed on resource limited embedded systems and are performance sensitive; others are deployed on cloud servers providing highly parallel services. Thus, SOA may provide effective solutions for managing the ever-increasing complexity while meeting the challenging requirements of CPS services on largely distributed, heterogeneous and dynamic resource environments.