Quality of service assesment using machine learning techniques for the NETCONF protocol

Abstract

Abstract: Study of an unsupervised machine learning approach for the testing results defined by the RFC2544 - ITU.Y1564 standard methodologies and the use of NETCONF protocol to automatically assess traffic parameters required to comply with quality of service level agreements. By doing disruptive and non-disruptive tests for service integrity, a service provider can certify that the working parameters of a delivered Ethernet circuit complies with the end user expectations, to avoid poor application performance. This work focus in an unsupervised learning approach using Expectation Maximization based clustering algorithm. We find that the unsupervised technique used is an excellent tool for exploring and classify service parameters like frame delay, frame delay variation, packet high loss intervals, availability and throughput. A correlation of parameters with the type of service required for the network flows (real time IP for data, video and voice applications) can be applied to automatically set bandwidth profiles. The bandwidth profiles can be configured per port, VLAN and CoS based, in one or multiple EVCs (Ethernet Virtual Circuits) per UNI device port. For the setup we adopt the Yang data modeling language and XML NETCONF message encoding protocol, followed by a delayed or an optional non-delayed orchestrated activation in the network devices via multiple NETCONF transactions.