Soutenance mémoire
SDN for WiFi Networks
SDN has been used for WiFi networks for implementing applications based service provisioning. Also, SDN is leveraged as a tool for implementing virtualization of WiFi access points (APs). Odin (Suresh et al., 2012) is a software-defined wireless network prototype for enterprise WLANs. It implements a flow-based virtualization technique to enable network operators to implement different WLAN services as network applications. In this architecture, an Odin master is the central controller entity that uses OpenFlow (McKeown et al., 2008) to program switches and APs that it controls. Each AP is packed with an Odin agent that communicates with Odin master by using Odin’s custom protocol. The applications on top of the Odin master uses Odin’s primitives to implement different enterprise services. Odin is a single operator solution to implement virtual AP abstraction and does not consider the case when multiple VNOs operate on a common infrastructure. Also, it does not consider abstraction and sharing of radio resources.
EmPOWER (Riggio et al., 2013) is an experimental testbed for SDN and NFV experimentation. The testbed’s data plane consists of OpenVSwitch (ope, c) and Click Modular Router (cli), while Floodlight (flo) has been used as the controller platform. It also utilizes a power management component called Arduino. This AP-based test-bed has provisions for implementing different network applications as slices. But the feasibility of implementing a resource allocation based multi-VNO platform is not discussed in the paper. Also virtualization of radio resources has also not been discussed.
SDN for Wireless Sensor Networks (WSNs)
To resolve the management problem of WSNs, SDN has been used for smart management of sensor networks. For example, a software-defined wireless sensor network (SD-WSN) (Luo et al.,2012) proposes a flexible, generalized architecture for WSN. To overcome the resource underutilization and network management problems of traditional application specific WSNs, the authors propose a programmable sensor network by following the control and data plane separation paradigm of SDN. To handle the data-centric characteristics of WSN, as opposed to the address-centric model of OpenFlow (McKeown et al., 2008), a modification of the Open-Flow protocol, named Sensor OpenFlow (SOF) has been proposed in this paper. SOF uses two different addressing schemes, Class-1: compact network-unique addresses and Class-2: concatenated attribute-value pairs (CAV) that suits the data-centric operation mode of WSNs.
Managing the control channel overhead and additional latency (due to data exchange between the control and the data planes) to ensure the desired performance SD-WSN would be a challenging task. Gante and et al. propose a WSN framework to facilitate management of a WSN (Gante et al.,2014). The authors propose a distributed control mechanism by incorporating a software-defined controller in each sensor BS. Application layer above the controller dictates the flowtable format of the sensor nodes. As dictated by the application (e.g., temperature, humidity sensing), the controller collects information from the sensor nodes and defines flow tables. For calculating the optimal routes among the sensor nodes, the controllers form an adjacency matrix that consists of the connection information (e.g., distance, signal strength, energy level,etc.)
SDN for Cellular Networks
A software-defined network paradigm has been proposed for cellular networks for both core and access network parts. These proposals leverage network programmability to foster rapid innovation, easier network management and also lower network CAPEX and OPEX. Some of such notable proposals are discussed in this section. SoftRAN (Gudipati et al., 2013) proposes a software-defined centralized control plane for radio access network. It abstracts all the base stations (BSs) in a geographical area as one virtual big base station, composing of a programmable central controller and individual base station function as radio elements. All cross radio element resource planning is made by the controller, i.e., if decisions of one BS impact the decisions of another neighbouring BS, those decision should be made by the controller. As the controller has a network-wide view, this scheme will help in reducing interference, smooth the handover process and also can facilitate data offloading. On the other hand, decisions that are based on frequently varying radio parameters should be taken locally by the individual radio elements. SoftRAN basically targets to ease the management of a RAN by providing better control on network management issues like: load balancing and interference management. SDMN (Pentikousis et al., 2013) is a SDN based implementation of cellular core networks.
It introduces a new MobileFlow stratum that decouples network control from the user plane. A MobileFlow controller controls the underlying MobileFlow forwarding engines (MFFEs) which are interconnected by IP/Ethernet network. MFFEs incorporate a standard mobile network tunnelling process, such as GTP-U, GRE encapsulation/decapsulation etc., that facilitate MobileFlow controllers to interoperate with legacy evolved packet core (EPC) nodes (e.g. MME, PGW, SGW, etc.).
SDN for heterogeneous Networks
OpenRoads (Yap et al., 2010a) is a seminal work on using SDN paradigm for wireless networks. This platform uses SDN to build a programmable virtualized wireless data plane. Open-Roads consists of basically three layers: a flow layer where the flow-tables of different data plane nodes are modified using OpenFlow (McKeown et al., 2008) protocol. Different wireless configuration parameters, like: SSID, wireless channel assignments, transmission power level are controlled and monitored by SNMP protocol. To enable resource sharing among multiple clients, a slicing layer is used to slice the network using the FlowVisor (Sherwood et al.,2009). The controller layer which is built on NOX (Gude et al., 2008), has a global view of the whole network and it allows the network applications (by different network users) to add/modify flow-table entries in the underlying data plane. OpenRoads is a heterogeneous platform that supports both WiFi and WiMAX networks. It has been shown that the platform supports seamless vertical handover between the disparate wireless technologies (Yap et al., 2010c). But the work does not discuss virtualization of radio resources (e.g., antenna, wireless spectrum, etc.).
Also the effect of elastic capacity provisioning in flow-based virtualization such as this, has not been studied in this work, which is a critical issue for an end-to-end virtual wireless network provisioning.
Virtualization without SDN
There have been works on wireless network virtualization that necessarily do not use the SDN concept of separating network control from the data plane. Network Virtualization Substrate (NVS) (R.Kokku et al., 2012a) is a WiMAX virtualization platform for creating virtual wireless networks on a common physical substrate. It is basically a MAC layer virtualization technique that allows bandwidth-based and resource-based slicing through a slice scheduler. Moreover it also incorporates customized flow scheduling for each slice in a BS.
A virtual base station architecture for WiMAX network is presented in (Bhanage et al., 2010b). In this model, virtual base stations are implemented in an external substrate that uses layer-2 switched data path and a control path to the BS. Radio resources of a BS is virtualized to create isolated slices that can implement different flow types with customized flow scheduling algorithms. SplitAP (Bhanage et al., 2010c) is a WLAN virtualization architecture, focused on fair sharing of uplink airtime across a group of users. A physical AP can be shared by different slices that can run different algorithms to control the UL airtime among different user groups.
In (Zaki et al., 2010b), the virtualization of the air interface of the LTE network has been studied. Here, a hypervisor was used for virtualizing the wireless spectrum.
Different experimental test-beds (using SDN or not) have been developed to do research on clean slate networking technologies leveraging virtualization. GENI (Bermana et al., 2014), Planetlab (Chun et al., 2003), AKARI (aka, 2009), SAVI (Kang et al., 2013), OFELIA (ofe), 4ward (Niebert et al., 2008) to name a few.
Programmable nodes and spectrum sharing
Shared access of radio nodes as well as wireless spectrum is critically important for virtualization of wireless networks. Virtual radio (Sachs and Baucke, 2008) is a virtualization framework that proposes to virtualize wireless nodes as well as the radio spectrum. In this model, the virtualization manager which is an InP-side component, takes virtual node instantiation requests from the prospective VNOs and upon the availability of resources creates new virtual nodes on a shared physical node. The paper however does not give any insight on how isolation would be managed among the incumbent VNOs that share a common physical node. Also the authors proposes to use various multiple access schemes (e.g., CDMA, TDMA, FDMA) for spectrum virtualization. But how to handle the added degree of complexity due to the virualization of radio spectrum is not discussed.
The spectrum virtualization layer (SVL) presented in (Tan et al., 2012a) is a sub-PHY layer that provides transparent abstraction for spectrum allocation. It allows dynamic spectrum al-location (DSA) to be implemented in a technology agnostic spectrum manager. SVL enables abstraction of the radio front-end which is very important for sharing (i.e., virtualizing) of the physical front-end by multiple players. One of the major advantages of SVL architecture is that it is fully implemented in software using the Sora (Tan et al., 2011) platform.
|
Table des matières
INTRODUCTION
CHAPTER 1 STATE OF THE ART
1.1 Wireless Network Virtualization
1.1.1 SDN for WiFi Networks
1.1.2 SDN for Wireless Sensor Networks (WSNs)
1.1.3 SDN for Cellular Networks
1.1.4 SDN for heterogeneous Networks
1.1.5 Virtualization without SDN
1.2 Programmable Radio Plane
1.2.1 Programmable Front-end
1.2.2 Programmable nodes and spectrum sharing
1.3 Cloud Computing for Wireless Networks
1.3.1 Cloud solution for Cellular Networks
1.3.2 Cloud solution for Heterogeneous Networks
1.4 Full Duplex (FD) systems
CHAPTER 2 DESIGN OPTIMIZATION OF WIRELESS ACCESS VIRTUALIZATION BASED ON COST & QOS TRADE-OFF UTILITY MAXIMIZATION
2.1 Introduction
2.2 Traditional Heterogeneous Network (HetNet)
2.2.1 HetNet dimensioning
2.2.2 HetNet cost analysis
2.2.3 LTE-TDD configuration
2.3 Locally Virtualized Network (LVN)
2.3.1 LVN architecture
2.3.2 LVN dimensioning
2.3.3 LVN cost analysis
2.4 Clustered/Remote Virtualized Network (CVN/RVN)
2.4.1 Network orchestrator (NO)
2.4.2 Radio access network (RAN)
2.4.3 Core network (CN)
2.4.4 CVN/RVN dimensioning
2.4.5 RRHs cost
2.4.6 CPC cost
2.4.7 Total CVN/RVN cost
2.5 Hybrid Virtualized Network (HVN)
2.6 Data Rate and Utility Function Construction
2.7 Results
2.7.1 Optimum CVN/RVN CPC size dcpc
2.7.2 CVN/RVN utility Uopt at different GP values
2.7.3 Optimum network utility Uopt of HVN for different GP values
2.7.4 Comparison of optimal network utility Uopt for different frameworks
2.7.5 Optimal CVN network coefficient popt c vs. cost weight wc and optimal CPC radius dopt cpc
2.8 Conclusions
CHAPTER 3 END-TO-END PROGRAMMABLE, CLOUD-BASED VIRTUALIZED HETNET: AN INTEGRATED ARCHITECTURE
3.1 Introduction
3.2 Requirements of Programmable Virtual Wireless Networks
3.2.1 Virtual Network (VN) Isolation
3.2.2 End-to-end Programmability
3.2.3 On demand resource provisioning
3.2.4 Network Function Virtualization (NFV)
3.2.5 Dynamic Spectrum Sharing
3.3 End-to-end programmable, elastic, HVWN
3.3.1 Management and Orchestration layer
3.3.2 Service layer
3.3.3 Application layer
3.3.4 SDN layer
3.3.5 Baseband signal processing
3.3.6 High capacity front-haul
3.3.7 Programmable virtualized radio nodes
3.3.8 Wireless spectrum virtualization
3.4 Business Cases for Programmable Heterogeneous Virtual Wireless Networks
3.4.1 Equipment manufacturers
3.4.2 Infrastructure providers (InPs)
3.4.3 Mobile Virtual Network Operators (MVNOs)/ Service Providers (SPs)
3.4.4 Third-party software companies
3.4.5 Inter ISP-VNO traffic offloading
3.5 Potential Research Issues & Challenges
3.5.1 Standardization of APIs
3.5.2 Balance between flexibility and complexity
3.5.3 Security threats minimization
3.5.4 Virtualization of wireless spectrum
3.5.5 Definition of isolation
3.5.6 Integration of Cognitive Radio (CR)
3.5.7 Backward compatibility
3.6 Conclusions
CHAPTER 4 SERVICE DIFFERENTIATION IN SOFTWARE DEFINED VIRTUAL HETEROGENEOUS WIRELESS NETWORKS
4.1 Introduction
4.2 Related work
4.3 HetNet Cloud architecture
4.3.1 Application layer
4.3.2 Software modules
4.3.3 Northbound API
4.3.4 NOS & East-Westbound API
4.3.5 Virtualization layer
4.3.6 Southbound API
4.3.7 InP’s resource management layer
4.3.7.1 Network Orchestrator (NO)
4.3.7.2 Resource broker (RB)
4.3.8 Baseband processing
4.3.9 Radio access plane
4.4 Using northbound API to facilitate virtual wireless network management
4.4.1 Interference management
4.4.2 Mobility management
4.4.3 Traffic offloading in a HetNet eco-system
4.4.4 Secured network
4.4.5 Internet of things (IoT)
4.5 Service differentiation in heterogeneous wireless networks
4.6 Challenges
4.7 Conclusions
CHAPTER 5 DEPLOYMENT OF FULL DUPLEX MULTI-CELL SYSTEMS FOR DENSE URBAN AND RURAL ENVIRONMENTS
5.1 Introduction
5.1.1 FD single cell deployment
5.1.2 FD multi-cell deployment
5.2 User selection and scheduling
5.2.1 Selecting users
5.2.2 Centralized Scheduling
5.2.3 Distributed scheduling
5.3 Dense Urban Model: Madrid Grid (MG)
5.3.1 Result analysis
5.3.1.1 User rate vs. SIC
5.3.1.2 System throughput vs. fairness
5.3.1.3 Node activity
5.4 Hexagonal grid (HG) model
5.4.1 Result analysis
5.4.1.1 User rate vs. SIC
5.4.1.2 System throughput vs fairness
5.4.1.3 Node activity
5.5 Conclusions
CONCLUSION
Télécharger le rapport complet
