CNS Core: Small: Design and Evaluation of Methods for Supporting Resilient and High-Availability Elastic Network Slicing

University Of Texas At Dallas, Richardson TX

Investigators

Abstract

Emerging 5G networks are expected to serve an array of critical applications in different industries such as healthcare, education, transportation, and manufacturing. A major challenge when supporting such a wide range of applications is being able to provide a high degree of availability and maintaining service requirements in the presence of network failures and other constantly changing network conditions. The concept of network slicing in 5G networks is widely viewed as a promising approach for supporting diverse service requirements by enabling the creation of application-specific network slices. A network slice consists of virtual computing and network resources that are tailored to the unique requirements of individual applications. Of particular interest is the concept of elastic network slicing in which the resources allocated to a slice may be dynamically adjusted over time in response to network failures or changes in network conditions. This project aims to address the issues of maintaining high availability and diverse service requirements in 5G networks through the intelligent design, provisioning, and restoration of elastic network slices. The work will result in techniques for providing a robust and resilient network infrastructure that is able support critical applications under dynamic and adverse conditions. The primary goals of this project are to develop methods and techniques for maintaining high availability and diverse service requirements for elastic network slices in 5G networks. To achieve these goals, this project considers a framework that utilizes techniques from deep reinforcement learning and online convex optimization to develop schemes for deploying, provisioning, reconfiguring, and restoring elastic network slices. Specific research problems that will be addressed include 1) machine learning based strategies for the pricing and admission control of elastic network slices while maintaining slice isolation and service requirements, 2) schemes for composing network slices and mapping slices to a physical infrastructure to optimize cost while providing availability guarantees in the presence of infrastructure failures, and 3) schemes for the progressive recovery of virtual and physical infrastructure components after catastrophic failure events. Methodologies will be developed in the context of online convex optimization that will provide a means to obtain theoretical performance bounds for a class of discrete optimization problems. The proposed techniques will be rigorously evaluated through comprehensive analytical models and simulation experiments. Testbed activities will be initiated to evaluate and validate proposed schemes. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.