Enhancing dynamism in management and network slice establishment through deep learning

With the variety of applications and the different user requirements, it is necessary to offer tailored resources efficiently not only in access but also in the core of the network. Inspired by the definition and standardization of mobile networks, especially 5G that focused on business verticals, the term network slicing has received numerous state-of-the-art efforts to materialize an approach that meets dynamism, programmability, and flexibility requirements. Leveraged by SDN and NFV technologies, network slicing is inspiring by resource sharing similar to virtual machine management, allowing standard network hardware to accommodate a wide variety of logical networks with specific requirements and data and control planes. However, state-of-the-art approaches do not address resource slicing at the core of the network in detail and appropriately. Therefore, we built NASOR to provide network slicing over the Internet data plane spanning across multiple domains through a segment routing and a distributed-based approach. Our approach excels those found in state-of-the-art by delivering an open policy interface that allows third-party applications to manage network slices dynamically. In this sense, this paper exploits this interface through a mechanism of convolutional neural networks that classifies network traffic, instructing the path-setting agent to be aware of application which predominantly runs on the network improving dynamism in the network slices deployment. Experiments showcase the convolutional neural network applicability and suitability as an enabling technology to enhance and instruct NASOR to establish network slices over multiple domains.