An effective prediction method for network state information in SD-WAN

Abstract

In a software-defined wide area network (SD-WAN), a logically centralized controller is responsible for computing and installing paths in order to transfer packets among geographically distributed locations and remote users. Accordingly, this would necessitate obtaining the global view and dynamic network state information (NSI) of the network. Therefore, the centralized controller periodically collects link-state information from each port of each switch at fixed time periods. While collecting NSI in short periods causes protocol overhead on the controller, collecting in longer periods leads to obtaining inaccurate NSI. In both cases, packet losses are inevitable, which is not preferred for quality of service (QoS). Packet loss needs to be reduced by minimizing the protocol overload on the controller and collecting accurate NSI to provide better QoS. This work proposes an effective prediction method for collecting NSI (PM-NSI) that significantly reduces packet loss and controller protocol load allowing the controller to collect accurate NSI in longer periods. The proposed method is compared against the existing NSI collection method, which collects NSI periodically, in use on the RYU controller and the Mininet emulator by using a dynamic routing algorithm. The test results indicated that PM-NSI reduces controller load around 1000% by collecting NSI in longer periods and so outperforms the existing periodic NSI collection method in terms of packet loss, jitter, controller load, and thus QoS.