Compression-based technique for SDN using sparse-representation dictionary

As Software-Defined Networks (SDN) emerged, the control and forwarding planes were abstracted using the standardized OpenFlow protocol which led to the increasing demand for optimal usage of the control link between the two planes especially for network monitoring. This paper proposes a data collection scheme based on a compression technique for SDN-based networks. It employs sparsity approximation algorithms for compressing the aggregated data in the SDN switch, while the recovery of the sparse data is taking place at the controller. The approach aims at further decreasing the link usage for Quality of Service (QoS) applications while increasing the network observability. The proposed solution extends the functionality of the SDN switch by integrating dictionary learning algorithms like K-SVD and Orthogonal Matching Pursuit (OMP) methods for the purpose of sparsity approximation. Experimental setup and the QoS link utilization metric for link monitoring were used for performance evaluation. The proposed solution was analysed over a range of sparsity levels, showing the data recovery accuracy of the controller under different compression ratios and using real internet traces. The results show that the proposed method reduces the control link overhead cost with up to 98% when compared to the case of periodic acquisition network monitoring of the SDN network.