| Abstract | Over the years the demand for large traffic volume and new advancements in network technologies have grown rapidly. The classical network in general has evolved towards complex architectures with vendor-specific designed interfaces. Unlike previous mobile network generations, 5G network provides a new foundational architecture with stringent requirements of network function virtualisation, massive scalability and high reliability with added flexibility. Alongside, the new concept of SDN-based network has offered potentially diverse benefits compared to the classical networks. Examples of the primary advantages include the centralized network provisioning and the network programmability. Therefore, Software Defined Network (SDN) with Network Functions Virtualization (NFV) have become a promising technique for the advanced 5th generation networks and the key components in the design of next generation networks.
As the Internet network is growing fast and operate wide range of traffic classes, the service quality and customer satisfaction are gaining importance in the network management. To this extend, policy-based network management becomes essential as certain traffic flows need to satisfy the business needs. In this research, a policy-based network management for quality services over SDN network is proposed. This research focuses on the traffic routing and measurement collection in order to satisfy the high level constraints. As part of the overall solution, the integration of intelligence in the framework is proposed. Machine Learning (ML) is becoming a very promising techniques to leverage the needs of autonomic and intelligent network management. This research proposes the use of Reinforcement Learning (RL) to enhance the decision making of policy-based network management for the end-to-end Quality of Service (QoS) guarantee. In this way, the proposed framework learns over time and it determines the best action to perform for ensuring end-to-end QoS delivery.
Recently the research on SDN has gained significant attention in the academia. In this context, the contributions of this research introduce novel methods to improve the monitoring and control towards intelligent traffic management solutions in multimedia-aware SDN-based environments. This research work brings three main contributions: (1) measurement collection and probabilistic-based routing solution is proposed to reduce the monitoring overhead over the control link between the forwarding and control layers in SDN, while increasing the observability of network state. This contribution applies a novel method based on sparsity approximation to compress the aggregated data in the SDN switch, while the SDN controller recovers the sparse data. Moreover, this research introduces an innovative probabilistic routing. The primary novelty of this contribution is the prediction of link bandwidth availability based on the Bayes’ theorem. In contrast to other studies, the proposed routing algorithm calculates the routing path when less information is advertised by the switch plane; (2) policy-based network management (PBNM) over SDN to enable QoS provisioning. It addresses the use of SDN centralized architecture to adopt the standard policy-based network management, while retaining the proprietary modular-block at each layer independently. With the help of policy management, the SDN controller can meet the requirement of an end-to-end service delivery with QoS guarantee; (3) reinforcement learning-based decision making for routing algorithms over SDN is proposed to apply a novel approach based on reinforcement learning method for the dynamic routing algorithm selection under SDN-based environments. Based on the learning approach, the proposed solution selects the most appropriate routing algorithm from a set of centralized routing algorithms that maintain the flow satisfaction with respect to the defined SLA requirements.
The proposed solutions were evaluated under diverse scenarios. In order to evaluate the applicability of the overall proposed system, several tools are used for the experiments: MATLAB, Mininet (SDN network emulator), Floodlight SDN Controller, OpenVSwitch/Ofsoftswitch13 (Software switch). While, the following tools were used for the traffic generation: VLC (Live video streaming), Ostinato (Network traffic generator tool), FFMPEG (record, convert and stream video). For the evaluation, the following experimental setup (Linux-based machine) was used: SDN controller and application: 2.2GHz multiprocessor of 4CPU unit, memory size of 16GB, Mininet network emulator: 2.2GHz of 4 CPU units, memory size of 32GB. The research presents the design and implementation of the framework that leverages the benefits of SDN and performance evaluation results are discussed to validate the feasibility of this approach. |
|---|