Address Aggregation in a Reverse-Learning Protocol

  • Forschungsthema:Routing Protocols, SDN
  • Typ:Masterarbeit
  • Betreuung:

    Polina Goltsman

  •  

    Background
    Currently, the IT industry is moving towards software-defined environments - the infrastructure is managed by software entities that react to current operating conditions in realtime. Part of this trend is adoption of Software-Defined Networks (SDNs) - a paradigm for network design where the control plane is separated from network devices and is implemented as a (logically) centralized controller. The controller is a software entity that runs on a general purpose computer. SDN is already being adopted in data-centers and there are currently several research projects aiming at adopting SDN for carrier networks including emerging 5G mobile networks.

    At ITM we are working on self-organizing SDNs -- networks where initial deployment and maintenance of SDN infrastructure is performed automatically. In this area we currently have following open theses.

     

    Problem Statement

    In SDN, each switch maintains a transport layer connection with one or several controllers. In the majority of current SDN deployments these connections run over a dedicated management network (separate cables). As this has several drawbacks, we consider SDNs with in-band control, where the switch-controller connections use the same network as normal traffic. In this case switches and controllers may be separated by several hops and a routing protocol is required to calculate paths and enable connectivity.

    We are currently developing a routing protocol that is optimized for maintaining connectivity between a large number of nodes and a small number of controllers. The proposed  protocol calculates multipath routes in the direction switch-controller while the routes in the direction controller-switch, which we call reverse path, are established using a reverse learning technique (as in MAC-learning). Additionally, we require that the protocol does not use manually assigned addresses. Currently, switches and controllers are addressed using flat unique addressing scheme which does not requre manual configuration, such as MAC addresses or cryptographic hashes which are unique with very high probability. In a naive implementation the addresses are not aggregatable and the forwarding tables of at least some switches will have to contain a large number of entries.

    The goal of this thesis is to develop solutions to reduce the size of the forwarding tables for the reverse path. Solutions such as label switching schemes, assignment of temporary aggregatable addresses, or similar, may be considered. The resulting scheme should be implemented and evaluated using a simulator.  The simulator is implemented in C++ for OmNet++ simulator library.

     

    Prior knowledge of C++ is helpful, but not required. The thesis (Ausarbeitung) can be written in either English or German.