Professor, Computer Science and Software Engineering
Network slicing constructs different logical networks (i.e., slices) on a unified physical infrastructure via Software-Defined Networking (SDN), network function virtualization (NFV), and cloud computing. Network slicing allows the support of highly diverse services, with possibly conflicting requirements, on a single infrastructure through an independent configuration of each network slice. Each slice in such an architecture is an end-to-end virtualized network instance and is tailored in terms of resources to meet the requirements of the services. The objective of the research problem is to design and manage a closed-loop mechanism to enforce the Quality of Service (QoS) throughout the lifecycle of each slice.
Such a closed-loop mechanism is made possible thanks to SDN, i.e., a network architecture approach that enables the network to be intelligently and centrally controlled, or programmed, using software applications. The closed-loop will therefore rely on SDN to easily and quickly manage, configure and (re-)optimize network resources with dynamic automated and proprietary-free programs, but also on big data and machine learning models and algorithms for the management of robust and reliable cognitive network slicing.
The closed-loop mechanism is required in order to achieve all that 5G wants to offers. A denser, fiber-rich network infrastructure needs to deliver the key performance indicators (KPIs): lower latency, higher data rates, ultra-high reliability and more connected devices. For that reason, the building blocks of the research project will be organized around the required KPIs.
Output of the projects should be a set of programs/software taking care of the automated definition of slices and of machine learning models and algorithms for the end-to-end QoS assurance and management in 5G slices.
The project investigates the research and development of static and dynamic autonomous network management agents that
collected datasets from EXFO network monitoring systems, we will research, analyze, design and develop machine
learning models and algorithms adapted to the analysis of time series. These algorithms can be used as part of
network monitoring toolsets to dynamically monitor the network behaviors. They will also allow real-time network
faults detection/prediction, performance anomalies, and cyber threats. Machine learning models and algorithms will
be tailored to assess the potential service impact of these faults, anomalies, and threats. This assessment will lead to
the recommendation of mitigations and solutions, along with their expected impact on the network. In non-real-time
contexts, we will analyze performance data collected over significant time periods and correlate them with the
evolution of the network in terms of variables, such as topology and configuration. This performance data analysis
will allow the identification of network areas with poor system performance and reliability.
© Concordia University