Machine-learning based routing pre-plan for SDN

Abstract

In Software Defined Network (SDN) environment, controller has to compute and install routing strategy for each new flow, leading to a lot of computation and communication burden in both controller and data planes. In this background, intelligent routing pre-design mechanism is regarded to be an important approach for routing efficiency enhancement. This paper investigates and proposes efficient SDN routing pre-design solution in three aspects: flow feature extraction, requirement prediction and route selection. First, we analyze and extract data packet and association features from user history data, apply these features into semi-supervised clustering algorithm for efficient data classification, analysis and feature extraction. After that, flow service requirement could be predicted through extraction of user, flow and data plane load features and implementation of supervised classification algorithm. Furthermore, we propose corresponding handling strategies related to data plane topology, flow forwarding and multi-constraint weight assignment, and proposes personalized routing selection mechanism.