Open Orchestration Cloud Radio Access Network (OOCRAN) Testbed

Open Orchestration Cloud Ra dio Access Network (OOCRAN) Testbed* Marti Floriach-Pigem Dept. of Signal Theory and Communication Barcelona Tech-UPC Castelldefels, Spain mfloriach90@gmail.com Guillem Xercavins-Torregrosa Dept. of Signal Theory and Communication Barcelona Tech-UPC Castelldefels, Spain guillemxercavins@gmail. com Vuk Marojevic Wireless@VT, Bradley Dep t. of Electrical and Comput er Engineering Virginia Tech Blacksburg, VA, USA maroje@vt.edu Antoni Gelonch-Bosch Dept. of Signal Theory and Communication Barcelona Tech-UPC Castelldefels, Spain antoni@tsc.upc.edu ABSTRACT T he 1 Cloud radio access ne twork (C-RAN) offers a revolution ary approach to cellular network deploy ment, management and evolution. Advances in software-defined radio (SDR) and networking techno logy, m oreover, enable delivering software- defined everything through the Cloud. Resources will be pooled and dynamically allocated leveraging abstraction, virtualization, and con solidation techniques; processes will be automate d using common application programming interfaces; and network functions and services will be programmatically provided through an orchestrat or . OOCRAN , oocran.dynu.com, is a software framework that is based on the NFV MANO architecture proposed by ETSI . It provi des an orchestration layer for the entire wireless infrastructure, i ncluding hardware, software, spectrum, fronthaul and backhaul. O OCRAN extends existing N FV manag ement fra meworks by incorporating the radio communications layers and their mana gement dependencies. The wireless infrastructure provider can then dynamically provision virtualized wireless networks to wireless service providers. The testbed ’s physical infrastructure i s built around a comput ing cluster that executes open-source SDR libraries and connects to SD R-based remote radio heads . We demonstrate the operation of OOCRAN and discuss the temporal implications of dynamic LTE small cell network deployments. CCS CONCEPTS • Networks → Wireless access points, base stat ions and infrastructure; Cloud computing; Network management * This is the author's version of the work . It is posted here for your personal use. Not for r edistribution. For citation purpos es, the definitive version of record is: M . Floriach-Pigem, G. Xercavi ns-Torregosa, V. Marojevic , A. Gelo nch- B osch, “Open orchestration Cloud radio access network (OOCRAN ) testbe d,” UCC'17 Compani on, Dec. 5 – 8, 2017, Austin, TX, USA . © 2017 Copyright is held by the owner/au thor(s). KEYWORDS Cloud radio access networ k; orchestration; lo ng-term e volution; network functions virtualization; testbed; software-defined radio 1 INTR ODUCTION We are witnessing an explosive expansion of Cloud computing and the use of data centers f or providing divers e types of commercial and non-commercial services. The benefits come from the increased freedom of development, scalabili ty and personalization of systems and services, and the rapid deployment of new functionalities, products and services. Virtualization is the enabler of resource sharing and its use reaches far beyond processing, storage and wired network ing [1]. Research and development (R&D) is now incorporating virtualization technology into wireless communications networks with a special emphasis on the physical layer using software-defined radio (SDR) technology [2 ] [3 ]. Wireless communications and Cloud computing have important di fferences that need to be considered when merging the two. The Cloud provides computing, storage, data, applications, and other services that are hosted on some remote physical resources, such as a data ce nter or a server connected to the Internet. Wireless communications networks have strict Quality of Service (QoS) requirements and exhibit a high degree of heterogeneity of equipment, syste m con figurations, and services. Moreover, currently deplo yed 4G lo ng-term evolution (LTE) networks fall short in terms of capacity and late ncy for enabling the tactile Internet, autonomous control of terrestrial and aerial vehicles, and massive connectivity of the Internet of Things. In addition, the end - to -end wireless network infrastructure and resources that deliver the services are often proprietary and heavily regulated. Despite the technological and non-technological barriers that exist today, usi ng and improving Cloud computing technology is n eeded to enab le the evolution of wireless communications and ne tworking te chnology and services [4 ]. The business model be hind t he Cloud radio access network (C-RAN) is based on the Infrastru cture- as -a-Service (IaaS ) model . A pool of physical resource s, which include antenna sites, networking components and proce ssors , is virtualized and offered to mobile virtual network operators (MVNOs) to build their networks as most suitable for the service s t hey intend to offer [5]. As technology e volves, MVNOs will be able to load software images of wireless networks that can be dynamically customized and adapted to t he changing operational conditions . That is, a MVNO can request more resources or release resources on the fly to “rei mage” i ts network. Whereas t his seems like a logical extension of Cloud computing, several research challenges remain. Telecommunications infras tructure providers ha ve been showing growing in terest in incorporating Cloud computing technology in t heir servi ce networks. The feasibility and benefits of using data centers a nd the Cloud le d to g radually moving their network infrastructure to a virtual environm ent. Durin g the 2013 Mobile World Congress several institutions agreed on standardizing the administration of the virtualized network infrastructure. An important p ush in t his direction has come from the European Telecommunications Standards Institute (ETSI) . ETSI clarifies the scope of net work fun ctions virtualization (NFV) and defines standard specifications that are mea nt to fulfill the operation al and management r equirements of next generation wireless ne tworks [ 6]. NFV decouples the phys ical network equipment from the network functi ons. A NFV implementation is understood as an i nstance, or virtual network function (V NF), and is complete ly software-defined. Re ference [7] demonstrates VNFs by executing the 3G and 4G core networks (CNs) in virtual machines (VMs). OpenStack and the kern el-based virtual machine (KVM) hypervisor provide the ove rall virtualization and management layers. Reference [ 8] presents the design, implementation, and evaluation of two LTE CN arch itectu res, one being based on t he principles of software -defined networking (SDN) and the other on NFV. Reference [ 9] describes the virtualization process of a base station and CN, whereas [ 10 ] performs a comparison between the proposed SDN and NFV solutions in mobile radio environments. Fig. 1 illustrates the ETSI NFV manag ement and orchestration (MANO) architecture , which i s composed of three main building blocks:  The Orchestrator manages the overall n etwork and i s responsible for i ncluding new servi ces and VNF packages.  The VNF Manager oversees the lifecy cle managemen t of VNFs on the NFV i nfrastructure (NFVI) according to the specifications provided by the orchestrator.  The Virtual ized Infrastructure Manager (VIM) manages the compute, storage and network resources of the NFVI. Figure 1: ETSI NFV MANO architecture. During the past ten years a plethora of Cloud computing solutions have been developed for providing IaaS. T hese tools allow managing servers as well as network infrastructure. Among the most popular IaaS frameworks are O penStack [ 11], Eucalyptus [ 12], and the Ubuntu Cloud infrastructure [ 13]. OpenStack has be come a prominent solut ion be cause of i ts capability to satisfy the network infrastructure providers’ needs in terms of massive computing and storage and complex networking. It can be considered as a VIM i n the ETSI MANO reference architecture. In our research we assume that wir eless infrastructure customers, t ypically MVNOs, lease resources as needed or periodically from an IaaS provider to build their networks [ 5] . These virtualized wireless networks will be tailored to the services they provide b y using virtual wireless ne twork components as thei r building blocks , rather than dealing directly with the computing, spectrum, remote radio heads (RRHs), antennas, fronthaul/backhaul, and other physical resources. The MVNO t hus deploys a net work of virtual resources with agreed communications, computing and networking capabilities. We have identified a gap between e xperiments in isolated radio environments and production -ready virtualized wireless networks. In order to leverage research and ed ucation in this field and facilitate transition t o practice, we introduce the Open Orchestration C-RAN (OOCRAN) testbed, o ocran.dynu.com . OOCRAN extends exi sting NFV management frameworks by incorporating the radio communications layers and some of their management dependencies. It provides an orchestration layer for the entire wireless infrastructure — hardware and software — so that the wire less infrastructure provider (WIP) can dynamica lly provision virtualized wireless networks to wireless service providers (WSPs) and satisfy t he i nstantaneous communi cations needs. This pape r presents our C-RAN testbed and illustrates the key features and example applications of OO CRAN. Section 2 summarizes the testbed highlights. Secti on 3 presents the OOCRAN archite cture and hardware components, where as Section 4 introduces the software layers. Section 5 discusses some use cases for deploying SDR LTE networks on the testbed to en able research and education. We conclude the paper with an outlook on re search on virtualized wireless networks, adaptive network services, and ultra-dense 5G networks. 2 TESTBE D OBJECTIV ES AND HIGHLI GHTS The testbed objectives were defined to support experimental research and educat ion on virtualized wireless networks in a laboratory and campus environment. These are:  Provi de a cloud computing platform for executing VNFs,  Enable real-time resource m anagement across heterogeneous resource pools,  Provide high flexibility, capacity, and extensibility, and  Be compliant with the ETSI NFV MANO architecture. The testbed de sign should (1) le verage the virtualization capability of C-RANs, as opposed to solely ce ntralizing the baseband processing, and (2) enable the generation and real- ti me processing of RF signals in a Cloud computing environment. OOCRAN uses the IaaS reference mode l as the basis of its design. It fi lls the orchestration and management gap of n ext generation v irtualized wireless ne tworks by providing a clear and easy way to dynamically con figure and assemble the building blocks of virtualized wireless networks and deploy them on shared infrastructure . It achieves this by virtualizing physical resources and managing shared access to limited computing and radio resources under a given policy. The modular design and isolation between hardware and software facilitates testbed upgrades. The OOCRAN framework manage s the virtual- to - physical resource mapping and allows creating new wireless networks as supported by the available ph ysical resources [ 14 ]. OOCRAN is an orchestration layer that follows the ETSI MANO architecture for creating, coo rdinating and managing wireless net works. An OOCRAN user can take the role of a WIP, a WSP or a wireless test provi der (WTP). This e nables analyzing different ways of splitting the resource deployment, management and maintenance re sponsibilities and e valuating resource access policies using a single framework. In the role of a WIP, O OCRAN can create complete communications syste ms by chaining several VNFs and delivering them to the WSP. Acting as a WSP, O OCRAN can precisely manage the wireless infrastructure and introduce p roper management policies to optimize the system behavior for t he given e nvironmental conditions and service demands. Actin g as a WTP, OO CRAN facilitates creating specific o perating conditions and test procedures and uses C loud -based system monitoring tools to analyze the system. In ot her words, a WTP creates lo cal or remote wireless laboratories that are tailored to the R&D needs. The O OCRAN framework has been developed using OpenStack Newton release as the VIM with the Neut ron and Heat op tions. The Neutron ex tension allows OOCRAN users t o customize remote access to the virtual networks they create. Heat, on t he other hand, allows saving a defined infrastructure as a description file, facilitating quick deployment or deletion. The KVM hypervisor provides an open source high performance platform. The core of the OOCRAN management environment is based on t he Django framework, which facil itate s the creation of complex, database- dr iven Web sites and thus simplifies the development of new use cases. The user interface is based on Web services, which allow activating schedulers, que ues, alarms , monitoring an d scripting tools to properly manage VN Fs a nd their configurations. These tools enable building a detailed control layer of the virtual infrastructure and applying various operational policies that take into account the state of resources and services. The wireless syste m i s implemented combining several VMs, each performing specific VNFs. These VMs execute SDR waveforms and run on general-purpose processors wit h access to SDR hardware and RF equipment. They are capable of generating and proce ssing real-time signals and interfacing radio links and CN functionalities. The currently available system is LTE, created as a fork of the srsLTE software library [ 15]. All the source code is open and released un der the AGPL li cense. It can be freely downloaded from the OOCRAN project repository [16 ]. The OOCRAN testbed is capable of ru nning in s imulation mode as well as creating an emulated or re al wireless network. A combination of simulated and real system components allows rapid prototyping and testing in controlled RF environments. 3 ARCHI TECTURE AND EQUIPMENT Fig. 2 depi cts the OOCRAN testbed architecture. It features several components that enable implementing, managing, and analyzing virtualized wireless networ ks. A computing cluster emulates the data center of the C-RAN. The RRHs are accessed through the radio aggregation unit (RAU) , using the terminology of the Next Generation Fronthaul Interface (NGFI) for 5G [ 17] . Together they form the remote radio system. RF instrume nts can be attached for signal or spectrum analysis, among others. Fig. 3a shows a photo of the testbed hardware. The roles and capabilities of the hardware compon ents are d iscussed in continuation. Figure 2: OOCRAN testbed architecture. CONTROLLER InfiniBand SWITCH COMPUTER1 147.83.1 18.228 Ethernet SWITCH External Equipment CONTROL DA T A RAU Remote Radio System COMPUTER2 Internet (a) (b) Figure 3: OO CRAN testbed hardware (a) and software (b) modules. 3.1 Computing Cluster The te stbed i s built around a computing cluster. Using the OpenStack te rminology, one PC acts as the Controller and the other two as Computers, all running Ubuntu 16.04. The Controller features a 3rd generation I ntel i7 8-core processor running at 2.5 GHz and using 8 GB of RAM. Its mission i s to administrate the C-RAN. Among other things it manages the VM lifecycle, user-defined network s, and virtual r outers. The t wo other Computers are two rackmount wor kstations, model SuperServer 6018TR-TF from SuperMicro. Each has two 2.6 GHz Intel Xeon 12 -core pro cessors, two Gigabit Ethernet ports , and one 56 Gbps InfiniBand port. These workstations host the VMs. They carry out the heavy computation, processing and forwarding the i ncoming data flows to the RF and CN components. The InfiniBand Switch IS5022 i s an 8 -port n on-blocking and unmanaged 40 Gbps swi tching system that is capable of delivering 640 Gbps bandwidth with 100 ns port- to -port latency. It acts as a gen eral switch of the testbed’s high speed ne twork that connects the c omputing cluster with the remote radio system. 3. 2 Re mote Rad io System One additional PC — Intel i7 -6700 8-core processor operating at 3.4 GHz with 32 GB of RAM — is included in the testbed as the RAU. It conne cts to the data center via the InfiniBand switch and to th e RRHs through Gigabit Ethernet . The RAU h andles the necessary data forwarding between the two networks. Five RRHs are currently available as part of the OOCRAN testbed. W e use the N210 model of Un iversal S oftware Radi o Peripherals (USRPs) from Ettus Research. These US RPs allow sampling at up to 50 Mega-samples per second (MS/s), are capable of g enerating or capturing RF signals below 6 GHz and conn ect to the processing center through Gigabit Ethernet. The remote radio system also features the following two RRHs: the ZedBoard w ith the AD-FMCOMMS3- EBZ daughterboard , capable of capturing 56 MHz of instantaneous RF bandwidth, and LimeSDR, which extends the bandwidth to 61.44 MHz. 3.3 E xternal Equ ipment RF instruments, such as spectrum analyze rs, as well as additional computing or radio equipment can be connected to the testbed through the Ethe rnet switch or the RAU. This allows extending the experimental and RF analysis capabilities of the testbed. 4 SOFTWARE LAYERS Th e OOCRAN software layers have been de signed to provide a wireless management framework that exten ds the functionalities of ETSI MANO. These layers facilitate t he focus on designing optimized management algorithms, called actuators . Our software framework, depi cted i n Fig. 3b, consists of six functional modules: GUI, MONITOR, NS_ENGINE, RF_ENGINE, QUEUE_TASKS and DRIVERS. 4.1 GUI The graphical user interface (GUI) provide s a user-friendly operating environment t o facilitate the i nteraction between the user (human network operator) and the vi rtual in frastructure or network serv ices (N Ss). The GUI has been developed using Python 2. 7 and Django. Djang o uses a mode l-view-controller, which facilitates making modification to the provided cod e . This allows saving the NSs and VNFs and defi ning actuators that can execute one or several tasks such as perform a partial reconfiguration or modify the lifecycle of the VNFs according to the state of the NS. 4.2 MONITOR The MONITOR module configures th e third-party programs Grafana/InfluxdB. InfluxDB is a framework that captures the state of the virtualized wireless infrastructure (VWI) and saves i t in a database. OOCRAN accesses this databa se, proc esses the saved states by searching for predefined patt erns, and starts the VWI reconfiguration process if the state matches the specified conditions. Grafana is a plotting tool that is used to pl ot the desired data. By using both f rameworks we are capable of capturing data from VNFs, processing the data and c reating customized graphs and alarms. This allow s processing and exposing t he state of a VNF (computational load, active users, waveform type, etc.). MONITOR captures the alarms that the Grafana/InfluxdB programs generate from the state of the NSs and VNFs. All alarms include a unique identifier related to a specific actuator. The module checks the credentials and the alarm i dentifier and, when both align, sends a command to NS_ENGINE to execute the corresponding actuator. This could, for example, trigger a partial reconfiguration of a set of VNFs. 4.3 NS_ENGINE The NS_ENGINE module manages the NSs and the actuators. It decides about the NS/VNF lifecycle or the actions following certain conditions (t ime, alarm, input, etc. ). This module is supported by RabbitMQ, which manages process queues and, among others, allows executing tasks asyn chronously. When the CON TROLLER VNF QUEUE_T ASKS MONITOR NS_ENGINE VNF VNF OpenStack RF_ENGINE GUI OOCRAN DRIVERS InfluxDB/ Grafana NS_ENGINE needs to apply a change to the infrastruc ture, it updates its own database and sends a new t ask to t he queue manager (QUEUE_TASKS). When the new task arrives, it is executed by the DRIVERS module that uses third -party application programming interfaces (APIs) to pe rform the update. 4.4 RF _ENGIN E The RF_ENGINE module manages the pool of radio reso urces (RF channels, transmi ssion power le vels, transmitter masks, etc.). It create s slices of spectrum and assigns them to different VNFs to avoid RF interference among coe xisting radios and networks. NS_ENGINE interacts with RF_ENGINE when creating a new NS or VWI. These modules are compatible wi th the VIMs of Ope nStack or Vagrant; thi s allows building and maintaining portable virtual software development platforms. 4.5 Inte rfaces The O OCRAN software base is a fork of the Django framework and, therefore, all Django APIs can be used. OOCRAN incorporates APIs from OpenStack, Vagrant , InfluxDB and Grafana t o create, delete, re configure and monitor the NSs. Fig. 3b shows the i nterfaces among the OOCRAN modules and other management components. The informati on exchanges between modules are done through calls to classes and th eir me thods. Third-party program drivers use HTTP RESTful APIs, whi ch allows installing RabbitMQ, InfluxDB, Grafana, and Open Stack on different computers. Alarms use Webhook, an HTTP callback that detects changes in the working conditions of a pr ogram. 5 EXA MPLE SCENARI OS The assessment of management frameworks or the d esign of resource management strategies can be done using simulated scenarios. On the ot her hand, the virtualization of physical resources and the associated management i ssues, such as resource slicing, isolation, and dynamic pro visioning, require real RF eq uipment. VMs can be effectively deployed to host LTE base stations ( eNodeBs) and user equipm ent (UEs). Additional VMs can simulate RF channels or connect to physi cal RRHs. This allows switching between simulated and real wirel ess links or using mixed links. Here we show the deployment of an LTE VWI and discuss VWI reconfiguration. 5.1 Virtualized LTE System The scenario fo r creating t he mini mum infrastructure for t he LTE downlink signaling is shown in Fig. 4. VNFs are attached to two different s ubnet s with different functionalities: a) the DataFlow Network carries the data instances and b) the Management Network carrie s the control instances. The IP address assignment i s done by means of DHCP in both cases. The DataFlow Network interconnects VNFs with external RRHs for sending or receiving IQ samples. The Management Network provides connectivity between OOCRAN/InfluxDB/SDN and the VNFs to send information Figure 4: Real LTE RF link setup. about operational states. More precisely, the state of the eNodeB VNFs is provided and VNF reconfiguration triggered using a secure shell (SSH) conn ection. The VNF application metrics are sent to the Influ xDB database using the Pyt hon API. Alarms from InfluxdB/Grafana are displayed at the following URL: http://oocran:8000/alerts/messages. OOCRAN assembles the required i nfrastructure building blocks for any real or simulated communications network and configures the corresponding VNF interfaces. The signal generated by the eNodeB transmitter instance is sent to the RRH. The radiated RF signal is captured by another RRH and sent to a spectrum a nalyzer instance. The spectrum analyzer instance e xecutes the UHD_FFT tool from GNU Radio (www.gnuradio.org). Fig. 5 captures the Horizon front end from OpenStack, the Ubuntu instance scree n and the UHD_FFT tool showing a capture of the LTE downlink spectrum gen erated by the eNode B instance. This spectrum exhibits en ough quality, that is, enough signal strength with respect to the noise floor. The UE recei ver, a UE instance with i ts USRP and processing unit, can demodulate the signal with a resulting block error rate below 0.1, which indicates proper system operation. 5.2 Dynamic VWI Deployment This scenario describes the dynamic deployment of a V WI , more precisely, a small cell -based wirele ss infrastructure (30 m cell radius). Here the OOCRAN management layer takes the role of a WIP. It desig ns and generates a suitable VWI on demand (using slices of physical resources) to satisfy the WSP needs. The virtual wireless network maps t o physical resources and includes slice s of RRHs, spe ctrum, t ransmission and processing power, among others. Some general assumptions of this scenario are  Use of omnidirectional antennas,  Line of sight links and free space path loss model,  RRHs can operate at various frequencies,  eNodeB sends data to multiple subscribers at t he same time and assigns different bit rates,  1.4 MHz LTE channel bandwidth. The si mulated LTE downlink scenario adds a simulated channel and UE su bstituting the RF path of Fig. 4. It includes several VNFs: a data source, an eNodeB transmitter, the channel, a UE receiver, and a controller that allows selecting the working parameters, such as channel conditions and LTE link parameters. eNodeB UDP :80 86 UDP :8 00 0 UDP :8 88 8 UDP, I P: 1 92 .1 68. 10 .x InfluxDB API SSH Data Flo w Netwo rk: 10 .0. 0.0 /24 Mana gemen t Ne twork: 1 92 .16 8.1 0. 11 - 254 InfluxDB OOCRA N Figure 5: Spectrum of the 1 .4 MHz LTE downlink signal that is transmitted by the eNodeB instance and captured by another RRH. Table 1: Time to setup a VWI eNodeBs Coverage Area Time 1  *( 30 m)² = 2826 m² 30.12 s 5 14,130 m² 33.49 s 10 28,260 m² 45.87 s 20 56,520 m² 60.19 s 30 84,780 m² 84.63 s Once the VWI i s con figured and de ployed for providing the specific servi ce, it needs to be periodically adapted to the changing operational conditio ns, including changing traffic loads, channel impairments, and service requirements. One possibility is to create a t ailored VWI for the new condition. The time that i s required to set up and deploy a new VWI needs to be considered to ensure non-disruptive NS. All subscri bers attached to the old VWI are re leased and remain disconnected from t he network until the setup of the new infrastructure is completed. The time it takes to set up and deploy a new VWI is a function of the number of eNodeBs needed to provide the desired coverage. Table 1 shows some figures. About a minute i s needed to deploy a new VWI on the Barcelona Tech/EETAC campus with an area of about 58,241 m². Fast swappi ng of a wor king VWI for another that better suits the new e xpected traffic load allows adapting the traffic capacity of the working VWI to the traffic demand. The relatively long times for VWI shutdown and redeploy ment calls for more sophisticated strategies to minimize the i mpact on user service perception. One solution i s defining lo ng periods for updates, e.g. one hour, or maintaini ng the ol d virtual networ k until the new network is established, introducing the n otion of VWI soft handover. Another solution is creating a VWI repository . OOCRAN can then sele ct the VWI that best matches the actual traffic demand, user di stribution, and ot her conditions or requirements. 6 CONCLU SIONS The di scussion on how to apply ETSI NFV MANO to manage practical C-RAN deployments is still in its early stage. This paper has introduced the OOCRAN testbed and a methodology for setting up a wireless access network with real and si mulated RF links. O OCRAN facilitates testing different infrastructure sharing methods and deployment strategies by providing monitoring and system analysis tools that do not jeopardize real - time execution. It enables creating and using a VWI repository for different types of e xperiments. The testbe d is modul ar and can be easily e xtended with hardware or software to take into account ad ditional environmental, service and network considerations, such as heterogeneous ne tworks. It pro vides a platform for experimental research and edu cation on virtualized wireless networks. Future work will address the creation of tailored and adapti ve network services , analyze dynamic VWI deployment solutions and funda mental limitations, and tackle ultra-dense 5G networks, where efficient orchestration is critical. AC KNOWLEDG MENTS This work has been partially supported by the Spanish Government, Mi nisterio de Ci encia e Innovación, through award number TEC2014-58341- C4 -3-R and the NLnet Foundation. REFERENCES [1] I. Gomez, V. Marojevic, and A. Gelonch . 2012. Resource Manageme nt for Software-Defined Radio Cl ouds. IEEE Micro , vol. 32, no. 1, pp. 44- 53, Jan.-F eb. 2012. DOI : 10.1109/MM.2011. 81 [2] I. Gomez-Mi guelez, V. Marojevic, and A. Gelonch. 2013. Deploymen t and Management of SDR Cloud Computing Resources: Problem Definition a nd Fundamental Limits. EURASIP Journ al on Wireless Com m. & N etworking , 2013:59, pp. 1-11, March 2013. [3 ] M. Richart, J. Baliosian, J. Serrat, and J- J Gorricho. 2016. Resource Slicing in Virtual Wireless Networks : A Survey. IEEE Trans. on Network and Service Management , Vol. 13, No. 3, pp, 462-476, Sept. 2016. [4 ] L. Doyle, J. Kibiłda, T. K. Forde, and L. DaSilva. 2014. Spectrum Without Bounds, Networks Without Borders. Proceed ings of the IEEE , vol. 102, no. 3, pp. 351-365, March 2014. doi: 10.1109/JPR OC.2014.2302743 [5 ] V. M arojevic, I. Gomez, and A. Ge lonch. 2012. Cognitive Resource Management f or all Wireless Access Layers . IE EE Ve hicular Tec hnology Magazine , Vol. 7, Iss. 2, pp. 100-106, June 2012. [6] ETSI-European Telec ommunications Standards Institute, http://www.etsi .org/ [7] Intel, White Paper, Carrier Cl oud Telecoms – Exploring the C hallenges of Deploying Virtualisation and SDN i n Telecoms Networks, https://www.intel.com/co ntent/dam/www /public/us/en/documents/w hite - papers/carrier-c loud-telecoms.pdf [8] A. J ain, N. S. Sadagopan, S. K. Lohani and M . Vutukuru. 2016. A Comparison of SDN and N FV for Re -designing the LTE Packet Core. In Pr oceedings 2016 IEEE Conf. on Network Func tion Virtualiza tion and Software Def ined Network s (NFV-SDN) , Palo Alto, CA, pp. 74-80. DO I : 10.1109/NFV-SDN .2016.7919479 [9] ETSI. 2013. N etwork Fu nctions V irtualisation (NFV); A rchitectural Framework . ETSI GS N FV 002 V1.1.1 , October, 2013. [ 10 ] C. Bouras, A. Kollia, and A. Papazois. 2017. SDN & N FV in 5G: Advancemen ts and challenges. I n Proceedings of the 2 0th Conference on Innov ations in Clouds, Internet and Networks (ICIN) , Paris, Fran ce, pp. 107-111. DOI: 10.1109/ICIN.2017.789 9398 [1 1] OpenStack - Open Source Softw are for Creating Pri vate and Pub lic Clouds Web Site, https://www.ope nstack.org/ [1 2] Eucalyptus Github Reposit ory, https://github. com/eucalyptus/eucalyptus [1 3] Ubuntu Documentation, UbuntuCloudInfrastructure, https://help.ubuntu.com/commu nity/UbuntuCloudInfras tructure [14] V. M arojevic, et al. 2012. Resource Management Implications and Strategies for SDR Clouds. J. Analog Integrated Circuits and Signal Pro cess in g , Vol. 73, Iss . 2, pp. 473-482, Sept. 2012, Springer US. [1 5] srsLTE Repository, https://gi thub.com/srsLTE/ srsLTE [1 6] OOCRAN Project Reposi tory, https://githu b.com/oocran [1 7] IEEE Standards Association, N GIF – Next Gener ation Fronthaul Interface Web Site, https://standards.ieee .org/develop/wg /NGFI.html