Demonstrating the Cost of Collecting In-Network Measurements for High-Speed VNFs

Abstract

Recent advances in the state-of-the-art of software packet processing along with the incarnation of SDN and NFV in networking brings the utility of software switches in production to a high level. Accompanied with the wide deployment of the latter, comes the practical and urgent need of monitoring networks that are composed of software forwarders/switches. On the one hand, this may provide new types of very finegrain operational data that can be collected, thus bringing the opportunity for network managers to get a deeper understanding of the underlying network state and performance. On the other hand, this massive data availability comes at a cost: software measurements can highly affect the measured values, thus biasing the collected data. The intensity of this bias becomes stronger when measurements are taken close to the data path. We believe that this trade-off should be explored more in detail, since the availability of fine-grained data offers new opportunities to apply machine learning techniques to infer changes in the network state, to forecast the evolution of some performance metrics or to automatically respond to event triggers without the human intervention. While our long-run objective1 is a full framework for performing automated test on software routing platforms, in this demonstration we focus on two key points that are prerequisite for our approach: (i) we showcase the impact of collecting the desired data within a Virtual Network Function and (ii) we setup a simple environment for data visualization on the same physical device.