?????? ????????? SDN ??????????????? ?????? BGP-LS ????????? ????????? ?????? ???????????? ?????? ???????????? ?????? ??? ??????

Software defined networking (SDN) is a new approach to design, to build, and to manage networks that separates the network's control and data forwarding planes to enable programmability. As the SDN is popularized in various networks range from access to core backbone, the SDN is extended to support multiple heterogeneous technologies. Moreover, in order to achieve scalability and confidentiality for the deployment of different carriers and domains, the network has to be divided into multiple domains. To provide end-to-end services across multi-domains, an efficient and scalable orchestration framework coordinates heterogeneous control planes and allocates the network resource across multiple domain. In multi-domain SDN networks, each domain has its own orchestrator that manages all the nodes and links via underlying SDN controller within the domain. To provide end-to-end services in multi-domain SDN networks, multiple orchestrators work together to coordinate heterogeneous domains and allocate the network resource. Orchestration framework wraps these orchestrators and provides a series of coordinating services of topology discovery, service provisioning, and management information, either distributed or hierarchical approach. In orchestration framework, topology discovery protocol is a key aspect of the framework because the end-to-end services highly depend on global network topology information. The topology discovery protocols can be distinguished into hierarchical and distributed mechanisms. In general, by using the distributed topology discovery protocol, each domain exchanges domain-specific network topology and reachability information with the neighboring domains. As a result, every domain maintains the global inter-domain topology information. However, the distributed exchanging of topology information requires complex interactions among orchestrators. It takes very long time for synchronization. On the other hand, hierarchical topology discovery protocols use a simple method to collect the topology information between parent orchestrator and child orchestrator, which avoids complex synchronization procedure. The parent orchestrator simply collects domain-specific topology information from all child orchestrators. However, only the parent orchestrator holds the global network topology. The determination of an end-to-end path across multiple domains can only be achieved by the parent orchestrator, since child orchestrators have no knowledge about other domains. This hierarchical approach causes a heavy burden at the parent orchestrator. In this thesis, we present a Global Topology Orchestration Protocol (GTOP), which is based on Hierarchical Distributed Topology Discovery (HDTD) protocol. This HDTD protocol combines the advantages of distributed and hierarchical protocols while avoiding their shortages. Using standard protocol instead of proprietary protocol that is an open solution without vendor lock-in. We implemented GTOP as an extension of BGP-LS that is a standardizing in IETF. The extended BGP-LS based GTOP, called GTOP-B, can be used as a topology discovery protocol for orchestrating multi-domain networks, since BGP-LS is designed to handle exchange of link-state information with traffic engineering capability, which includes link, node, and prefix properties and attributes. In order to validate the GTOP-B protocol, we implemented and compared it to the other conventional approaches (e.g., hierarchical and distributed approaches). Our result shows that hierarchical distributed approach could solve the slow convergence time of distributed approach, and make per-domain path computation and provisioning, while reducing traffic concentration on parent orchestrator in hierarchical approach. Moreover, we validated that BGP-LS based GTOP-B protocol has high interoperability, compatibility, extensibility and low-burden. Most of the BGP-LS Network Layer Reachability Information (NLRI) subtypes could be interoperable between multi-vendors. For the compatibility aspect, the attributes of BGP-LS conclude all the attributes of other routing protocols (e.g. OSPF, IS-IS, RIP). Likewise, BGP-LS is a kind of concise protocol without any redundant data while keeping the accuracy of topology information. In addition, this protocol can be further extended to the legacy networks as well as newly invented network.Maste