Scalable Optimized Design of Multi-layered Network Slicing and Virtual Network Embedding Under Long Range Dependent Traffic and Flexi-grid Optical Transmission with Adaptive Modulation

Description

Network slicing has emerged as a promising technique to realize provision of specialized and dedicated virtual networks, called slices, to provide efficient networking solutions that support their customers’ QoS requirements. Slices are built on a common physical network infrastructure using network function virtualization (NFV) and software defined networking (SDN) technologies. Multiple layers are needed because of the complexity and heterogeneity of internet traffic, customers and their QoS requirements. Telecommunications networks are traditionally layered (e.g. OSI model). Modern and future wireless networks are expected to simultaneously support a multitude of diverse mobile applications including IoT applications that require services at multiple networking layers. The internet customers themselves are layered where a higher layer customer further virtualizes its own virtual network to smaller layered virtual networks. Multi-layered network slicing can be viewed as a generalization of multi-layered virtual network embedding (MVNE) over optical networks where a network slice is a set of networking (wireless and wireline), storage and/or computational resources.  This project will provide a software tool that includes novel algorithms for efficient multi-layered network slicing and virtualization. I key novel feature of the algorithms is their general nature that enable considerations of under realistic traffic conditions and many technologies, e.g. elastic optical networks (EONs). Specific novel features include:(i) the consideration of Variable Bit Rate (VBR) traffic modeled by Long Range Dependent (LRD) processes and statistical multiplexing, (ii) MVNE over WDM or Elastic Optical Network (EON) with an arbitrary set of layers and realistic traffic models, (iii) MVNE over EON (MVNE/EON) with adaptive modulation, and (iv) asymptotically optical algorithms for layered network slicing. We will build on ideas from our earlier work on the multilayer market algorithm (MMA) (which has polynomial complexity), and results from reinforcement learning and multi-armed bandit problems for which asymmetrically optimal solutions are proved only for a very special case for a single layer. This approach will lead to solutions that meet customer QoS requirements in multilayeredMVNE and network slicing. We seek to maximize earnings before interest and tax (EBIT) with consideration to OPEX and amortized CAPEX, network resource utilization effects, and service disruption penalties. Heuristic algorithms that generate provable bounds on the quality of solutions, and that scale to large instances beyond those handled by integer linear programming (ILP) will be provided for validation. Our tool with have a friendly user interface for network inputting, editing and visualization to extend our industry and education impact.