Programmable RAN Sharing in 5G systems
Sujet proposé par
Directeur de thèse:
Unité de recherche UMR 7102 Laboratoire de recherche d'EURECOM
Domaine: Sciences et technologies de l'information et de la communication
Vision Existing cellular systems, such as LTE, are built upon expensive proprietary equipment and complex control-plane protocols, and thus they do not offer enough flexibility to program the network, and quickly adapt to the network condition. The network openness, flexibility and scalability come from the proper abstraction of low-layer functions in a hierarchical manner. Such abstraction and control approaches are required in 5G radio access networks so that we can orchestrate the network to a ubiquitous and unified service platform. The proper abstraction hides details of the low-layer implementation while providing only the necessary data for information exchange and high layer control and coordination. This will bring multiple key advantages to 5G networks, in that it will: • manage the complexity and greatly simplify the signaling for implementing advanced MAC/PHY layer technologies; • extend self-organizing network features in 5G as the abstraction enables low-layer control entities to be controlled and coordinated; • offer network-wide programmable control plane for rapid creation and deployment of new services, for which the abstraction is the essential step for high-level programmability. Therefore our vision on control and coordination in 5G radio access networks is a simple, open, unified and programmable control framework powered by proper abstraction of low-layer states, behaviors and functions, common control protocols, interfaces and primitives for network coordination and automation, and hierarchy abstraction on network states for efficient resource allocation and spectrum management in the network.
Main Objectives In the last years, much attention (both from Industry and Academia) has been placed on software-defined networking (SDN). In one of its most widely acknowledged definition, SDN essentially consists of two elements: (a) a clear separation of control and data plane, and (b) a well-defined interface or abstraction between control and data plane. This decoupling separates the network intelligence from the infrastructure whose complexities are abstracted away from the control applications. Nevertheless, the underlying SDN concepts are not novel per-se, i.e. software has always been used to control the network and control and data plane have been separated in several network architectures (e.g. in optical networks). The fundamental innovation introduced by SDN essentially consists in a vendor- and programming language-agnostic interfaces to networking gear. While SDN indeed reduces part of the intrinsic complexities of the current networking architectures, its actual embodiments pay little attention to the requirements of the radio access domain. As a matter of fact applying the tools currently available in the SDN ecosystem, e.g. OpenFlow, would essentially result in treating mobile terminal as hosts wired to an Ethernet switch. In a radio access network, however, the concept of a “link”, due to the stochastic nature of the wireless medium, is loosely defined, e.g. allocating a flow at a certain BS can affect the available bandwidth at another BS. The definition of new abstraction methodologies is then of capital importance in order to enable true programmability of the RAN. To this end, this thesis will focus on the Programmability of Radio Access Network (RAN) considering the RAN sharing use case. Two innovative objectives have been identified: • Physical and MAC layer modelling and abstraction to provide a simple network view of low-layer reality and available resources, and thus to enable a scalable and flexible control and coordination framework for complex resource coordination in 5G networks. • Programmable control based on the low-layer abstraction and machine learning techniques with well-defined open interfaces and protocols to greatly simplify the management of heterogeneous mobile networks, to be verified by efficient resource coordination algorithms developed for identified 5G use cases. Radio Access Network Sharing is a mechanism to dynamically allocate and share spectrum (frequency band), radio resources (i.e. time, frequency) and physical resources (e.g. compute, network, storage) among different tenants while maintaining their respective performance isolation. We will explore system trade-offs for a programmable RAN according to the Consistency Availability and Partition tolerance (CAP) conjecture, which states that it is impossible for any distributed system to provide at the same time the following guarantees: consistency, availability, and partition tolerance. In particular, consistency in policy application, network availability and partition tolerance in the face of fading channels, multi-path loss and mobility have not been studied in the wireless sphere, while in the networks literature only some first steps have been taken. As such there is no suitable theoretical model for dealing with the CAP conjecture in a programmable RAN. In addition we plan to study network virtualization and multi-tenancy, namely topology and resource abstractions and performance isolation. The former aims at defining the formalism through which tenants, i.e. the owners of the virtual networks to be instantiated on top of the physical substrate, are sharing resources and specifying their requests. Such request may involve a certain network topology in terms of nodes and links as well as performance constraints in terms for example of expected bandwidth and/or interference. The latter aims at investigating how to assign to each slice its own fraction of network resources (spectrum, radio, hardware), and how a traffic allocation decision, taken inside a slice, can affect the other slices (both logically and from the performance standpoint).