Novel Adaptive Data Collection based on a Confidence Index in SDN

Abstract

SDN makes networks programmable by bringing flexibility in their control and management. An SDN controller decides how packets should be forwarded and installs flow rules on the data-plane devices for this end. For efficient control and management, the SDN controller needs to monitor the network state continuously in order to have an accurate and up-todate view of the underlying data-plane. The need for this highvisibility on the network brings specific types of monitoring as streaming telemetry where data is streamed continuously from network devices (wired switches, wireless energy constrained nodes, etc.) as bulk time series data. But this may generate a lot of overhead. Adaptive monitoring techniques provide ways to reduce this overhead, but generally, they require complex user parameter tuning and also they effectively handle data dissemination overhead with counterpart certain energy-consuming treatment on the source nodes. In light of this, we propose novel adaptive sampling technique based on a confidence index that considerably reduces the number of exchanged messages about 55-70 % while maintaining an accurately collected data, represented by the explained variance score that is about 0.7 and 0.8. And more, our proposition achieves these results while being a lightweight solution for source nodes.

Low-Overhead Near-Real-Time Flow Statistics Collection in SDN

“…Our proposition is an easy to deploy, transparent and lightweight solution that comes from our early previous work [8], where we propose, a generic adaptive algorithm based on a confidence index for collecting any type of information (CPU usage, switch information, etc). Here we leverage this adaptive algorithm that is enhanced and optimized for statistics presenting a particular pattern, an increasing trend (counters).…”