EAGER: Generalized Network Slicing for NSF OpenCloud

Stanford University, Stanford CA

Investigators

Abstract

This project will demonstrate the feasibility of generalized network slicing for Software Defined Networks (SDN) that would enable multiple multi-tenant virtual clouds on the same physical infrastructure. The generalized network slicing is aimed at network operators as opposed to the end users or tenants of a cloud. The network operators require network slicing with the ability to specify performance, topology, network programmability, and independent address space and require performance isolation. In this regard the proposed network slicing is distinct compared to other efforts in the broader area of network virtualization and can be used independent of network virtualization aimed at supporting 100s and 1000s of cloud tenants. The project will demonstrate generalized network slicing system by focusing on four fundamental challenges: - Topology Mapping. Goal is to allow an experimenter or a tenant to specify topology of her network slice and map the (virtual) slice topology on the physical network and maintain this mapping as both physical and slice topologies change over time. - Address Space Mapping. Goal is to allow a tenant to use her own private address space and translate the private address space to the physical address space while exploiting packet forwarding capabilities of the existing devices. - Control Function Mapping. Goal is to allow each tenant to have SDN style programming capability of her virtual network slice and map control functions on to the corresponding physical instantiation of the network. - Performance Isolation. Goal is to allow each tenant to specify resource requirements for her network slice and allocate and police resource usage for each virtual slice to ensure performance isolation. The project will focus on demonstrating the functionality and gaining valuable experience with experimentation. This will help the community to subsequently focus on performance, scale, robustness, and other attributes of the solution that are needed for a real world deployment. The cloud is rapidly changing the face of computing, and as such, is having an increasing effect on experimental systems research, teaching and research in the broader Computer Science community, and how research is done across the breadth of science and engineering. By demonstrating the feasibility of network slicing as an important building block of a research cloud this project will inform the the on-going discussion about the cloud?s evolution.