Intelligent Resource Orchest ration for Service Function Chaining and Reactive Traffic Steering in Software Defined Network

Abstract

In the network softwarization, Network Function Virtualisation (NFV) has shifted the standard of network services deployment and management for telecommunication. However, the allocation of physical resources to the Virtualised Network Functions (VNFs) dynamically, efficiently, and autonomously is one of the key tasks to achieve this objective. Network designing is a serious factor for the deployment of VNFs to create a Service Function Chaining (SFC) with efficient resource utilization and optimal path in the edge computing environment. The development of a self-driven Software Defined Networking (SDN) also requires an intelligent network design, especially to find optimum routing patterns for traffic steering that meet the goals set by administrators. But unfortunately, the current network designing methods do not fulfill the desired requirement for precise estimations of related performance metrics. Recently, the application of Artificial Intelligence (AI) is considered by the research community to operate and control the network. The Graph Neural Network (GNN) is adopted as an AI solution in the field of the network because GNN can understand the complex relationship between network traffic features, routing, and topology to produce an accurate estimation of relevant performance metrics. In this thesis, we propose the implementation of the Knowledge Defined Networking (KDN) system using well-known open-source technologies. GNN is incorporated into the system to predict the optimal path for SFC deployment and traffic steering. Two major applications; an SFC deployment application and a Traffic Steering application are developed for the knowledge plane to achieve SFC deployment in the edge clouds and traffic steering. An algorithm path selection module is developed to assist the SFC deployment application while deciding the best path selection and a policy configurator module is developed to translate a selected path into an SFC for the Open Source MANO (OSM). Similarly, a path selection module is developed to assist traffic steering applications while deciding the best path selection and a policy creation module is developed to translate a selected path into a flow rule policy for the Open Network Operating System (ONOS). Both applications interact with the GNN model for predicting the Key Performance Indicator (KPI) of paths in the network and provide the optimal path for SFC as well as traffic steering using the algorithm mentioned above. Most importantly, the OSM policy configurator and ONOS policy creation module reduce human error while configuring a desired service on the network. Monitoring modules are developed for both applications to fetch the network statistics after a few intervals of time and store them in the database which is required by the GNN module for path prediction. Lastly, ONOS and OSM clients are added in both applications to establish the communication between ONOS and OSM respectively. Another application, Traffic Flow Manager (TFM) is developed for the ONOS controller to perform as a broker between the knowledge plane and control plane. TFM is not only a broker but also performs real-time monitoring for network topology and deploy flow rules in network devices using the SDN controller. With the help of results, the proposed system is evaluated using the complete virtualized environment which is composed of ONOS, OpenStack, OSM. During the evaluation of the proposed system, we have achieved efficient resource utilization for SFC deployment and maximum throughput while steering traffic in the distributed edge clouds based on the optimal path selection considering less latency among the source and destination. Also, the proposed system reduces CPU utilization's load while steering the network traffic in a dynamic network topology. As compared to the existing state of the art SFC deployment in the edge cloud and traffic steering system, the proposed system is improved efficient resource utilization for SFC deployment up to 20%. Also, maximum throughput up to 5% and CPU load up to 13% while steering traffic in the distributed edge clouds.