SDN-based Platform Enabling Intelligent Routing within Transit Autonomous System Networks

Nowadays, huge volume of traffic is generated and transported over Wide Area Networks (WAN). WAN is composed of multiple transit Autonomous Systems (ASs). Scaling traditional WAN to deal with the new application’s traffic profile and to fulfill the requested Quality of Service (QoS) requirements becomes too expensive and complicated due to physical resource limitations. In view of the above facts, in this paper we propose and implement a Software Defined Network (SDN) based WAN (SD-WAN) platform enabling the development of new generation of routing algorithms (e.g., using operational research or artificial intelligence techniques). The design makes use of several popular open-source projects such as ONOS, Docker, Mininet, Iperf and Quagga for building the emulated SD-WAN topology. We develop an external management and resolver unit with Python that (i) collects periodically useful statistics such as packet loss rate and instantaneous throughput of the transit AS links. (ii) Permits to compute route path according to some algorithm for every new flow arrival from a neighbor AS to another one. (iii) Configures the data plan according to the computed path. A network operator can implement any routing algorithm respecting the application requirements (e.g., a predefined packet loss threshold) or improving its revenue. The topology size and the traffic profile are easily changeable. For demonstration purpose of the platform usability and functionality we implement a simple shortest path routing (SPR) algorithm and leave the development of intelligent routing algorithms to future works. As expected, results from a simple SPR implementation show that SPR yields to poor performance because of the under-use of all the available resources (e.g., links not included in the shortest paths), making space for improvement and optimization.