This paper presents the experiences of building ONOS, an experimental distributed SDN control platform designed to meet the performance, scalability, and availability requirements of large operator networks. The authors describe and evaluate two ONOS prototypes. The first prototype implemented core features such as a distributed, logically centralized global network view, scale-out, and fault tolerance. The second prototype focused on improving performance, particularly event latency. Based on these prototypes, the authors identify additional steps needed for ONOS to support use cases such as core network traffic engineering and scheduling, and to become a usable open source, distributed network OS platform that the SDN community can build upon. ONOS adopts a distributed architecture for high availability and scale-out. It provides a global network view to applications, which is logically centralized even though it is physically distributed across multiple servers. ONOS follows in the footsteps of previous closed-source distributed SDN controllers such as Onix, but it is intended to be released as an open source project that the SDN community can examine, evaluate, extend, and contribute to. The first prototype faced several challenges, including consistency and integrity issues, low performance and visibility, data model issues, excessive data store operations, and polling. These issues led to the development of the second prototype, which focused on improving performance by introducing an event notification framework, optimizing the data model, and implementing a topology cache. The second prototype significantly improved performance, reducing latency for network state changes and reaction to network events. However, it still did not meet the target throughput for path setup. The authors also demonstrated ONOS on the Internet2 network, showing its ability to operate on a real WAN, using virtualized hardware and software switches, and demonstrating faster ONOS and link failover. The authors discuss future work, including improving isolation and security, exploring new ways to parallelize and distribute large workloads, and preparing a usable open source release of ONOS. They also plan to explore three broad use cases: traffic engineering and scheduling of packet optical core networks, SDN control of next generation service provider central offices and points of presence, and remote network management. The paper concludes with a discussion on the potential of ONOS as an open, distributed network OS that can meet the performance and reliability requirements of large production networks.This paper presents the experiences of building ONOS, an experimental distributed SDN control platform designed to meet the performance, scalability, and availability requirements of large operator networks. The authors describe and evaluate two ONOS prototypes. The first prototype implemented core features such as a distributed, logically centralized global network view, scale-out, and fault tolerance. The second prototype focused on improving performance, particularly event latency. Based on these prototypes, the authors identify additional steps needed for ONOS to support use cases such as core network traffic engineering and scheduling, and to become a usable open source, distributed network OS platform that the SDN community can build upon. ONOS adopts a distributed architecture for high availability and scale-out. It provides a global network view to applications, which is logically centralized even though it is physically distributed across multiple servers. ONOS follows in the footsteps of previous closed-source distributed SDN controllers such as Onix, but it is intended to be released as an open source project that the SDN community can examine, evaluate, extend, and contribute to. The first prototype faced several challenges, including consistency and integrity issues, low performance and visibility, data model issues, excessive data store operations, and polling. These issues led to the development of the second prototype, which focused on improving performance by introducing an event notification framework, optimizing the data model, and implementing a topology cache. The second prototype significantly improved performance, reducing latency for network state changes and reaction to network events. However, it still did not meet the target throughput for path setup. The authors also demonstrated ONOS on the Internet2 network, showing its ability to operate on a real WAN, using virtualized hardware and software switches, and demonstrating faster ONOS and link failover. The authors discuss future work, including improving isolation and security, exploring new ways to parallelize and distribute large workloads, and preparing a usable open source release of ONOS. They also plan to explore three broad use cases: traffic engineering and scheduling of packet optical core networks, SDN control of next generation service provider central offices and points of presence, and remote network management. The paper concludes with a discussion on the potential of ONOS as an open, distributed network OS that can meet the performance and reliability requirements of large production networks.