Wireless Pers Commun DOI 10.1007/s11277-017-4285-1
Augmented RAN with SDN Orchestration of Multitenant Base Stations Salvatore Costanzo1 Lazaros Merakos1
•
Dionysis Xenakis1 • Nikos Passas1
•
Springer Science+Business Media New York 2017
Abstract Network virtualization (NV) has been successfully applied in wired networks, providing abstraction of the networking equipment and simplifying the network/resource management procedures. However, more light needs to be shed on how the emerging NV technologies, including software defined networking and network function virtualization, can be used by the mobile network operators to efficiently handle the ever-increasing demand for mobile data traffic. In this paper, we focus on the long term evolution system and present a NV-based framework that aims at bringing the access network closer to the end user. The proposed framework enables the dynamic sharing of base stations and radio resources among different mobile network operators, a.k.a. multi-tenant operation. Accordingly, we present novel NV triggering and decision algorithms to demonstrate how the proposed dynamic sharing of base stations can be deployed based on predefined service level agreements. Extensive system-level simulations accompany the paper, showing that, even with the use of simple NV triggering and decision algorithms, the proposed framework results in notable performance improvements at the users of the home operator, i.e. the sending operator, without significantly affecting the performance of the users at the host operator, i.e. the operator that shares its infrastructure. Keywords Network virtualization Software defined networking Radio access network Multi-tenancy 5G
1 Introduction The unprecedented explosion in the number of communication-enabled mobile terminals, requiring higher throughput, lower application-layer delay, and reduced energy consumption, has recently launched the discussion for architectural innovations towards the
& Salvatore Costanzo
[email protected] 1
Department of Informatics and Telecommunications, University of Athens, Athens, Greece
123
S. Costanzo et al.
fifth generation communication network system, a.k.a. 5G [1]. 5G is foreseen to support communication scenarios with massive yet diverse mobile applications and services, requiring ubiquitous connectivity to the radio access network (RAN). Satisfying this demand leads to a bifurcation of the RAN system that enables network densification, i.e. a more dense deployment of cells for increasing the RAN capacity. On the one hand, network densification is considered as a straightforward solution for boosting the area spectral efficiency and improve network capacity [2], but on the other hand, it comes at the cost of increased monetary and network management overheads owing to the increased number of small-sized base stations involved. Besides, the use of dedicated radio resources, or networking equipment, hinders the employment of flexible and spectrumagile resource management, further increasing the capital and operational expenses required for the mobile network operation, i.e. CAPEX and OPEX, respectively. For that reason, network operators need to investigate the architectural enhancements that will enable them to satisfy the requirements set for the 5G system in a cost effective manner. To this end, network sharing has recently attracted a surge of interest by the network operators as it is envisaged as a promising approach for maximizing the effectiveness of the resource utilization stage and boost the return on capital, e.g. by enabling more efficient sharing of physical network components, including base stations or core network elements. The Third Generation Partnership Project (3GPP) has also recognized the importance of network sharing and has provided a suite of requirements and guidelines [3, 4] for enabling the sharing of the RAN among operators that have defined a priori service level agreements (SLAs). Besides, as the implementation aspects of the 3GPP RAN sharing procedures are left open, the research community is paying a lot of attention in devising solutions for enabling effective network sharing. To this end, the main trend in the recent state-of-the-art is to employ the concept of network virtualization (NV) [5] that aims a simplifying the network management process in network sharing scenarios. NV separates the physical network infrastructure from the network management process, allowing the coexistence of multiple networks over the same physical networking infrastructure. Even though NV has been initially deployed in fixed/wired networks, showing enhanced networking efficiency, increased resource utilization, and lower CAPEX/OPEX, the key principles of NV have recently driven the design of the future mobile RAN [6]. In the context of a RAN, NV enables the dynamic sharing of network resources/equipment among different mobile network operators, further improving network capacity and lowering the costs required for network maintenance. Software defined networking (SDN) [7] and network function virtualization (NFV) [8] are two of the most promising NV-based technologies that are envisaged to be integral part of the future network architecture. SDN provides a clear architectural separation of the network control and forwarding functions enabling network operator to migrate intelligence from the edge networking nodes to a (logically) centralized software based entity, a.k.a. SDN Controller, capable of maintaining a global view of the network status. By applying this approach, the physical infrastructure is abstracted for applications and services and the network control is directly programmable by the SDN Controller in a purely logical software fashion. This approach facilitates on-the-fly adaptation of the network behavior to different scenarios, by re-programming the software running at the SDN Controller, enabling the network to be more flexible and respond quickly to changing traffic requirements. NFV is a complementary technology to SDN that aims at simplifying network management operations bringing cost efficiencies, time-tomarket improvements and speeding up the innovation in the networking. More specifically NFV deals with the implementation of specific network functions in software so as to enable migration from proprietary equipment to general-purpose hardware platforms. NFV
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
employs virtualization techniques to deploy network functions in virtual servers, e.g. data centers, allowing the support of faster innovation cycles in the network through a simple software update (rather than hardware replacement). Notably, the combination of these technologies will play a key role in migrating from the today’s heterogeneous network architectures to the integrated 5G architecture of the future, i.e. the transition phase beyond 4G. The aim of this paper is to investigate how the aforementioned NV-based technologies can be smoothly integrated in the current 3GPP architecture in order to support dynamic sharing of RAN resources among multiple operators. In this direction, we propose a framework that integrates SDN and NFV into the 3GPP RAN sharing architecture, with the aim to realize network elasticity in the RAN sharing process, referred to as multi-tenant operation, adapting it to the real time status of the network. The proposed framework aims to dynamically increase the RAN capacity of a particular mobile network operator, termed as the tenant operator, by augmenting its available RAN resources on demand. RAN augmentation is performed by virtually increasing the base station density of the tenant operator allowing it to lease on-the-fly the physical networking infrastructure of another third-party infrastructure provider, i.e. another multi-tenant operator. The aforementioned process can be performed based on a priori SLAs established by the mobile network operators involved in the multi-tenant operation, while it can take place under certain conditions that relate with the network status of the involved operators, e.g. when additional base stations are required to offload traffic, or when link quality improvements can be achieved. The contributions of this paper include: • A NV-based framework that leverages the benefits of NV, NFV and SDN for enabling operators to improve their RAN capabilities on demand under specific circumstances. The proposed framework includes an enhanced architecture for multi-tenant RAN and a logical signaling flow for the proposed augmented RAN operation. • A comprehensive discussion on the architectural and functional enhancements required to implement the proposed framework in the existing 3GPP Long Term Evolution (LTE)/LTE-Advanced (LTE-A) network and a detailed specification of the signaling procedure. • Extensive system-level simulation results assessing the performance gains following from the proposed architecture as compared to the legacy 3GPP LTE/LTE-A architecture. The remainder of the paper is organized as follows. In Sect. 2, we summarize the related works. In Sect. 3, we describe the proposed framework along with a comprehensive discussion on the signaling flow required to support it in the LTE/LTE-A system. In Sect. 4, we present extensive system-level simulations to assess the performance of the proposed framework under different scenarios, whereas in Sect. 5 we draw our conclusions and provide insights for future work.
2 Related Works RAN and spectrum sharing can significantly reduce capital investment for future 5G system, e.g. by reducing the number of active network equipment, e.g. base stations, for setting up the RAN infrastructure. Recently, 3GPP has overviewed a suite of service and
123
S. Costanzo et al.
business requirements for enabling the RAN sharing paradigm in LTE networks [3, 4]. More specifically, 3GPP Services Working Group (WG) SA1 analyzed a set of use cases for RAN sharing, while 3GPP WG SA2 specified the architecture and procedure to enable different operators to share the RAN and proposed two different approaches for RAN sharing, named multi operator core network (MOCN) and gateway core network (GWCN) respectively. In the MOCN approach each operator owns an independent core network while the RAN base stations are shared with other operators. In the GWCN scenario the operators share also the mobility management entity (MME). From one side, the GWCN approach can enable extra cost saving as compared to the MOCN approach; however the GWCN approach is less flexible as it doesn’t support internetworking with legacy networks and mobility among multiple radio access technologies (Multi-RAT). MOCN is the most used approach in the current state of the art and is also the reference approach under the scope of this paper. Current literature includes different approaches for resource sharing; some of them propose to leverage the benefits of NV to enable the creation of virtual base stations from physical base stations that are shared among multiple operators. In [9] authors propose a simple model for base station sharing among multiple operators in LTE, based on creating a group of logically independent virtual base stations, referred to as Virtual eNodeBs (VeNBs). A key feature of the VeNBs is that they can be operated by different operators at the same time. Although simulation results demonstrate improvements in terms of load balancing at the base stations, more light needs to be shed on the adaptability, flexibility, and elasticity of the proposed model in the future 5G network. Authors in [6] propose a RAN sharing technique named Network Virtualization Substrate (NVS) to enable flexible spectrum sharing among multiple operators. NVS is defined as a hierarchical scheduler consisting of two layers. At the first layer the scheduling operation is controlled by each operator in an independent way assuring isolation, while the second layer consists of a flow scheduling layer that allocates resource to each operator taking into account at-priori SLA agreements. This technique enables greater flexibility in slicing the resources in both uplink and downlink directions. Similar to [6] the authors in [10] propose a procedure that enables operators to deploy independent resource scheduling policies among the VeNBs instances running in a shared LTE base station. To achieve this, they employ a dynamic two-layer resource scheduler composed by (1) a common physical scheduler, which is responsible for frequency domain scheduling, and (2) a set of virtual schedulers, each of which gives the VeNBs the freedom to implement customizable scheduling policies. In another work [11], the authors focus on the MAC functions required for allocating the resources within a shared evolved NodeB (eNB) among multi-tenant operators. To achieve this, they introduce an entity called Hypervisor that is responsible for allocating the available physical resource blocks (PRBs) among multiple operators in line with predefined SLA agreements. Different from [6, 9–11], in this paper we focus on the procedures to achieve a flexible allocation of the resources in a shared RAN by leveraging the benefits of SDN, NV and NFV technologies. These technologies are envisaged by the 5G public private partnership (5G-PPP) [12] as the enabling tools for accelerating service innovation towards more efficient and sustainable architectures for 5G. In this direction, a plethora of research works have started to investigate SDN solutions with the aim to simplify and improve the efficiency of the network management process. In our previous work [13] we propose a framework that permits the virtualization of LTE eNBs in a more flexible way by leveraging the benefits of SDN. In this framework the virtualization of the eNBs is dynamically handled in an SDN fashion, enabling one operator to offload traffic on-demand to the base
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
stations owned by other operators. In another work [14] we propose an SDN-based framework for enabling elastic spectrum sharing in multi-operator environment of Frequency Division Duplex (FDD) macro eNBs and Time Division Duplex (TDD) pico-eNBs. The aim of this framework is to enhance flexibility and the resource management efficiency by enabling an SDN-based coordination of network resource management process. Different from our previous works [13, 14], in this paper we focus more on the architectural enhancements and on the 3GPP compatibility issues of an SDN-based approach for RAN sharing. The potential benefits of applying SDN in the LTE network have been also discussed in [15, 16], where the authors investigate a scenario wherein all the control plane functionalities at the LTE core network (CN) nodes are migrated to a logical centralized SDN controller. Moreover, all the management plane entities are replaced by virtual machines that are orchestrated in an SDN fashion. Authors indicated that such an SDNbased architecture reduce the signaling message overhead between the control plane entities as compared to the legacy 3GPP LTE architecture. In [17] authors propose an SDN-based LTE-Evolved Packet Core (LTE-EPC) architecture that employs the OpenFlow protocol [18], i.e. a control protocol that interfaces SDN controllers with the SDNbased network elements, for managing all the control and data plane procedures at both serving gateway (S-GW) and packet gateway (P-GW) nodes. The control plane of these entities and the MME functions are packaged together as applications on top of an SDN controller. Authors have shown that the proposed OpenFlow-based procedures are capable to lower the signaling load overhead as compared to the legacy control plane procedures. Other works investigate the novel concept of Cloud-Ran (C-RAN) [19] that aims at enabling energy efficient network operation and cost savings on baseband resources. The key idea behind C-RAN is to simplify the base station (BS) architecture by moving all the base-band (BB) processing in a central physical location, while leaving a certain number of antennas at the so-called Remote Radio Heads (RRHs). A key advantage of this solution is that the aggregation of multiple BB units into a single one reduces the number of base station sites and lowers the operation costs required. In [20] authors propose a virtualized radio access layer that abstracts the base stations functions into a cloud base station. The physical base stations are transformed in simple radio elements that are remotely controlled by a logical centralized controller. The aim of this architecture is to enable infrastructure sharing among different RAN technologies. C-RAN is considered as the main candidate for the next generation 5G architecture, however even though the benefits of employing the C-RAN architecture are clear, a smooth transition between prominent 3GPP systems (e.g. LTE-A) and forward-thinking 5G architectures is of critical importance. Aiming to cover this critical divide and make the most out of the recent advances in SDN, in this paper we present a framework specific for multi-tenant scenarios that aims at enabling flexible sharing of the RAN resources among multiple operators in a less invasive manner, i.e. by means of a simple software update, and provide a clear evolutionary step of the LTE/LTAA architecture towards 5G. Different from current state-of-the-art solutions, this paper contributes to the following important research domains: • Extends the 3GPP specifications for RAN sharing with and SDN-based management approach that aims at facilitating the execution of network sharing decisions by taking advantage of the global network view offered by the SDN paradigm. • Achieves more flexibility in the network sharing process by exploiting the underutilized resources in the multi-tenant shared RAN to enhance the network capacity of the tenant operators.
123
S. Costanzo et al.
• Provides valuable insights on how to introduce non-invasive architectural changes in the current LTE/LTE-A architecture to enable effective RAN-sharing operations based on SDN. • Provides extensive numerical evaluations measuring the cost/benefits of the RAN sharing operation, evaluating both the user performances and the signaling cost resulting from the employment of NV and SDN in the 3GPP LTE/LTE-A network.
3 Proposed Multi-tenant Architecture In this section we introduce the proposed framework, which consists of an enhanced architecture for multi-tenant RAN and a SDN-based procedure for dynamic RAN-sharing. Firstly, in Sect. 3.1, we present an enhanced architecture for multi-tenant RAN and we discuss its compliance with the recent 3GPP specifications for RAN sharing. In Sect. 3.2 we outline the signaling procedure required to implement the proposed framework in the LTE/LTE-A system. The key functions of the proposed framework are overviewed in Sect. 3.3, where we additionally provide exemplary triggering and decision algorithms to control the dynamic sharing of base stations among two operators, while in Sect. 3.4 we provide a preliminary evaluation of the he signaling load of the proposed procedure for RAN sharing.
3.1 Enhanced Architecture for Multi-tenant RAN 3GPP specifications [3, 4] have defined the role of the actors involved in the RAN sharing process. In a RAN sharing scenario, the main actor is the master operator provider (MOP), that deploy a set of base stations endowed with sharing capabilities, referred to as multitenant base stations, forming the so-called shared-RAN (S-RAN). The MOP shares the S-RAN with multiple operators, a.k.a. participating operator providers (POPs). It is assumed that the POPs own their RAN, i.e. they already serve their users with own base stations, but they are allowed to lease the RAN infrastructure of the MOP under certain circumstances, e.g. for improving coverage or for offload traffic in case of traffic congestion. In a typical RAN sharing scenario the MOP is willing to the share its RAN with other POPs with the aim to maximize the utilization of its network resources and make a profit by the leasing operation. At the same time the POPs are willing to lease the RAN infrastructure of the MOP as they can increase their coverage capacity without the need to install additional base stations and incur in additional CAPEX/COPEX. Taking into account these specifications, we propose an enhanced architecture for multitenant RAN, that leverages the benefits of SDN and NFV technologies to enable dynamic sharing of the RAN resources and simplifying the management operations in a multi-tenant scenario. The enhanced architecture is depicted in Fig. 1. In the enhanced architecture the RAN and CN nodes perform the same tasks and use the same interfaces as in the legacy 3GPP LTE-EPC architecture, referred to as baseline in the sequel. As in the baseline version, the eNBs provide user and control plane protocol terminations towards the users (UEs) and are interfaced with the MME and S-GW by the legacy S1-MME and S1-U interfaces respectively. The S-GW handles the user plane, as it is responsible for routing and forwarding the user data packets towards the P-GW, and the P-GW is responsible for inter-connecting the remainder RAN elements to the Internet. However, different from the baseline version, the P-GW and S-GW are replaced by SDN-aware switches wherein all
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
Operator A Home operator (POP)
EPC A S-GW Ctrl RIB
P-GW MME Ctrl
Operator B Host operator (MOP)
EPC B MT-B
MT-A
MME
INTERNET
SDN MC-A
P-GW Ctrl
S-GW Ctrl
SDN MC-B
RIB
OAM
OAM
P-GW B
P-GW A
S-GW A
S-GW B
non-shared eNB B eNB A
eNB A
L3 Tunnel S5 SGi SDN Southbound API (incapsulates S1-MME) SDN Northbound API (Itf-N) SDN East/West API
OpeNB
eNB A
UE A
S-RAN Operator B
Augmented RAN (for Operator A)
OpeNB
VeNB A
S1-U Virtual S1-U
PDCP
PDCP
RLC
RLC
MAC
MAC
VeNB B
non-shared eNB B
LEGEND
NFV API Hypervisor Layer Abstracted PHY Layer
Fig. 1 Proposed multi-tenant RAN architecture
the control functions are separated from the data forwarding functions and moved in controller applications (P-GW Ctrl and S-GW Ctrl in Fig. 1) running in a logic centralized entity referred to as SDN Main Controller (MC). Moreover all the control plan protocols are replaced by an SDN control plane protocol like OpenFlow [18], as proposed in [17]. The MME, that is responsible for handling the control plane functions for mobility management throughout the mobile operator’s network, is executed as an application running at the MC. In order to achieve compliance with the 3GPP standards, the messages sent at the S1-MME interface are encapsulated into SDN control messages and forwarded via SDN Southbound interfaces [7] to the MME controller application running at the MC. Accordingly, a mapping function running at the MME controller application, maps the 3GPP control plane messages in SDN control plane messages as proposed in [17]. Another controller application running at the MC is the Multi-Tenant Manager (MT) that is responsible for handling the multi-tenant operations. Moreover, the reaming of 3GPP control plane operations (not explicitly depicted in Fig. 1) are also running as controller applications at the MC. Let us now to focus on the RAN sharing scenario under the scope of this paper. Without loss of generality, we consider the presence of two mobile network operators, referred to as home operator (operator A) and host operator (operator B) respectively. We assume that the operator A acts as a POP and owns only a set of legacy base stations (eNBs A in Fig. 1). The operator B acts as a MOP and deploys a set of legacy eNBs forming a nonshared RAN together with a set of multi-tenant base stations, referred to as Open eNBs (OpeNBs), forming the so-called S-RAN. The OpeNBs are assumed capable of sharing their physical resources among multiple operators in a dynamic fashion, as described in our previous work [13]. Nevertheless, the sharing process is based on SLAs that are established between operators A and B prior to the sharing of the physical resources between the two operators. Under this viewpoint, we consider that operator A can lease the base stations of the S-RAN owned by operator B, i.e. the multi-tenant OpeNBs, and create an augmented
123
S. Costanzo et al.
RAN on demand by performing handovers (HOs) towards the cellular base station (independent of the operator) that can better satisfy the user requirements. Notable, the enhanced architecture in Fig. 1 fits one of the of high practical interest use case scenario analyzed in the 3GPP specification [4], named ‘‘Single DM for managing S-RAN and POP own RAN’’. The reference architecture for this use case scenario is depicted in Fig. 2. In this scenario [4], one operator (Operator A in Fig. 2) manages its own (non-shared) network elements (NEs), i.e. eNBs and RAN resources, from its RAN domain manager (DM), referred to as POP-RAN-DM. A second operator (Operator B in Fig. 2) acts as the Operator B in the enhanced architecture in Fig. 1 and manages a set of base stations, endowed with sharing capabilities (S-RAN), together with a set of legacy base stations. The network manager of the POP (POP-NM) cooperates with the network manager of the MOP (MOP-NM) in order to manage the operations for RAN sharing according to SLAs established among the two operators. As 3GPP does not specify the implementation details of such a network sharing process, this paper aims to address this gap, by providing a set of architectural enhancement and procedures for the aforementioned use case scenario. More specifically, we propose: • A specific architecture for the base stations forming the S-RAN, which we call OpeNBs. • An SDN-based signaling procedure that enables the communication between the POPNM and the MOP-NM (highlighted boxes in Fig. 2), with the aim to facilitate the execution of network sharing decisions by taking advantage of the global network view offered by the SDN paradigm. Aiming to clarify the role of the entities present in the enhanced architecture for RANsharing, in the remainder of this section we describe and motivate our architectural choices in more details. The key features of the OpeNBs and the MC at the different mobile operators respectively are summarized in Table 1, while a more detailed description of their functionalities is provided in Sects. 3.1.1 and 3.1.2 respectively.
POP -NM
POP -NM MOP -NM
I-N POP – RAN-DM
EM-Shared RAN
EM- non Shared
Single DM
eNodeBs
S-RAN
Operator A Fig. 2 Reference shared RAN architecture for 3GPP [4]
123
Operator B non-shared RAN Operator B
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations Table 1 Key features of the proposed network entities OpeNB
MC host operator
MC home operator
Enables the sharing of the available resources among multiple operators Provides access to UEs registered to other operators Virtualizes the base station infrastructure in a pool of emulated eNBs (VeNBs) Allocates a VeNB instance to each home operator
Assists the resource sharing operation at the OpeNBs Coordinates instantiation of the VeNBs at the OpeNB Handles tunnel connection between the OpeNB and the CN of the home operator Monitors the SLAs
Creates on-demand augmented RAN Takes handover decision towards the OpeNBs Configures the tunnel endpoint at CN of the home operator
3.1.1 OpeNB An OpeNB is capable of supporting all functions supported by a legacy eNB but also acts as a multi-tenant base station. To achieve this, the OpeNBs adopts a slightly modified protocol stack compared to the one supported by the legacy eNB. In particular, we consider that the physical (PHY) layer of an OpeNB provides similar functionality with the one provided by the PHY of a legacy eNB. Nevertheless, we also consider that they implement an additional layer, termed as the Hypervisor Layer, which runs on the top of the PHY layer. The purpose of this layer is to abstract the PHY from the upper layers of the protocol stack. To achieve this, the Hypervisor Layer virtualizes the physical resources and allocates them to a pool of virtual software instances that we term as VeNBs. Each VeNB emulates the behavior of the remainder upper layers of the protocol stack, while it is logically connected to the CN of the appropriate operator by means of a Layer 3 (L3) tunnel link. In more detail, we consider that the Hypervisor Layer is the place where the NFV technology is hosted, towards achieving the virtualization of the physical base station infrastructure. At this point, we note that a typical NFV-based architecture is composed by numerous building blocks of virtualized network functions (VNFs) [8] that run on top of the physical network infrastructure. In our architecture, these VNFs are software instances that are responsible for handling specific network functions, including the functionalities of the upper layers in the eNB protocol stack. The VNFs are considered to be chained together in a building block fashion, with the aim to deliver a full networking service, e.g. to emulate the behavior of each VeNB. Besides, the Hypervisor Layer handles the logic interface between the PHY layer functions and the VNFs that emulate the upper layers, by means of NFV application programming interfaces (NFV APIs) [8]. In such operations, the Hypervisor Layer is assisted by the MC that lies within the CN of the host operator, e.g. the MC-B in Fig. 1.
3.1.2 Main Controller The MC is a logical central entity that is responsible for the control plane management of the enhanced architecture including the management of the multi-tenant operation. The control plane operations are managed at the MME, S-GW Ctrl and P-GW Ctrl controller applications, while the multi-tenant operation is handled by the MT controller application (MT-A and MT-B in Fig. 2). The MC is capable to acquire a global knowledge of the
123
S. Costanzo et al.
network state by interacting with the legacy 3GPP Operations Administration and Maintenance (OAM) system via the Eastbound API [6], being able to retrieve the status of a set of Key Performance Indicators (KPIs), such as throughput, SINR, etc. Such information feeds the RAN Information Base (RIB) [4] facilitating a global network view to the SDN controller. We consider that the MC, which lies at the CN of the home operator (MCHome) acts as POP-NM and exploits the acquired knowledge to perform mobility management operations in a SDN fashion. On the other hand, we consider that the MC of the host operator (MC-Host) acts as MOP-NM and utilizes its knowledge to handle the resource sharing operation at the OpeNBs. More specifically, the MT controller application running at the MC-Home (MT-Home) is responsible for taking the HO decisions towards an OpeNB of the host operator. In the sequel, we term such decisions as NV Decisions. Note that during a NV Decision, the MT-Home identifies the most appropriate target base station among the OpeNBs of the host operator (see Sect. 3.3.2 for more details). To this end we consider that MT-Home maintains a logic channel with the MT controller application running at the MC-Host (MT-Host) to exchange information about the status of the OpeNBs of the host operator, in a real-time fashion. From the management perspective this channel is based on the 3GPP Eastbound API that acts as the interface between the MOPNM and the POP-NM. The information exchanged between the MOP-NM and POP-NM is also used by the two MCs to establish a tunnel connection between the OpeNB of the host operator and a CN entity of the home operator. Aiming to clarify the tunnel establishment phase, let us assume that a tagged user of the home operator (UE A in Fig. 1) is served by the OpeNB of the host operator (Fig. 1) as result of a successful HO. As mentioned in Sect. 3.1.1, at this point the VeNB that refers to the home operator should be logically inter-connected to the CN of the home operator, e.g. by means of an L3 tunnel. Such a tunnel is required for providing a logical interface between the control/data plane protocols running, on the one hand, at the VeNB instance at the OpeNB, and, on the other hand, at an appropriate attachment point in the CN of the home operator, e.g. the MME/S-GW of the serving eNB. Since this procedure should be transparent to the CN entities of the home operator, we consider that the tunnel establishment and maintenance is performed by the MC-Home, which is additionally responsible for forwarding the control/data plane packets from the MME/S-GW at the home operator to the target VeNB, in an OpenFlow fashion [18]. Let us now focus on the MC that lies at the host operator, i.e. MC-B in Fig. 1. Among other tasks (e.g. the management of the SDN-based control plane), the MC-B is responsible for handling the requests for a NV-HO, i.e. the HO toward the OpeNBs of the host operator, sent by the home operator via the Eastbound Interface. Such incoming requests are then processed by the MT-Host controller application running at the MC-B. The MTHost performs admission control and verifies whether the incoming requests are in line with the pre-established SLAs (see Sect. 3.2 for more details). Once the HO requests are accepted, the MT-Host provides all necessary functionality to handle the sharing of the OpeNB’s resources among multiple VeNBs. Moreover the MT-Host assists the Hypervisor Layer at the OpeNBs in configuring the logic connection between the UEs served by an OpeNB and the appropriate VeNB instance. Finally, the MT-Host configures the logical connection, i.e. L3 tunneling, between each VeNB instance and the respective CN of the home operator.
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
3.2 Signaling Flow In Fig. 3 we summarize the signaling flow of the proposed procedure for RAN sharing. The signaling flow below is performed for the first time when a UE registered to Operator A is transferred to an OpeNB that is owned by the Operator B (see Fig. 1). The first two steps are integral part of the baseline HO procedure. The serving eNB sends to the tagged UE a ‘‘Measurement Configuration’’ message, specifying the set of physical measurements that need to be performed (step 1). In the following, the UE reports to the serving eNB the results of the respective measurements (step 2) and, based on the reported measurements, the serving eNB decides whether a HO towards an OpeNB is required or not (step 3). We refer to this type of handovers as NV-HO in the sequel. The triggering of a NV-HO can be based on different types of criteria that relate with the current network status and can be reprogrammed at the eNBs by the corresponding MT controller application (MT-Home). In Sect. 3.3, we provide some exemplary NV triggering algorithms based on the signal quality at the UE or the cell load at the eNBs of Operator A. If the triggering criteria for a NV-HO are met, the serving eNB forwards all necessary information on the respective UE (including the derived measurements) to the MT-Home (step 4) via the legacy 3GPP Southbound Interface. Accordingly, based on its global knowledge on the network status, the MT-Home (MT-A in Fig. 3) identifies the OpeNB of the host operator (operator B) that satisfies a predefined set of NV decision criteria (step 5). We refer to this phase as the NV Decision phase (see Sect. 3.3.2) in the sequel. Note that MT-Home has only access to a
Home Operator A S-GW Ctrl
UE A
Serving eNB S-GW P-GW
Host Operator B P-GW Ctrl
MME
MT-B
MT-A
MME
SDN MC-B
SDN MC-A
1-Measurement Config 2Measurement Report
3-NV Triggering Algorithm
4-Report UE for NV HO (UE A)
14- P-GW TUNNEL Fwd Rule
P-GW S-GW Ctrl Ctrl
6OpeNB Access 5-NV decision Request (Op A, OpeNB X, UE A)
7NV SLA monitor
13OpeNB Access ACK + TUNNEL Establishment Info
P-GW
S-GW
OpeNB
8-OpeNB Access Request 9-NV Admission Control
10- OpeNB Access Ack 11- S-GW TUNNEL Fwd Rule 12- P-GW TUNNEL Fwd Rule
15- S-GW TUNNEL Fwd Rule 16- L3 TUNNEL Establishment 19-HO Command
18- NV HO ACK (UEA,OpeNB)
17- NV HO ACK ( UE A,eNB A,OpeNB)
20-NV HO confirmation
Fig. 3 Proposed NV signaling procedure
123
S. Costanzo et al.
restrict set of information regarding the status of the OpeNBs, e.g. traffic load, location, etc. This information is shared by the host operator (i.e. the MOP) with the home operator (i.e. the POP) on demand with respect to the pre-established SLAs. Once the NV decision is taken, the MT-Home sends an OpeNB Access Request to the MT-Host (step 6), containing the identity of the home operator (Operator A), the identity of the target OpeNB, as well as any information related to the characteristics of the ongoing UE connections. The identity of the home operator is required at the host operator (operator B) to verify whether the NV-HO request is in line with SLA agreements between the two operators (step 7). On the other hand, the identity of the target OpeNB is needed at the MT-Host to forward the OpeNB Access Request towards the respective OpeNB (step 8). Upon receiving the OpeNB Access Request message, the target OpeNB utilizes the information on the characteristics of the ongoing UE connections (included in step 6) in order to infer on whether it can resourcefully host the connections of respective UE (step 9). This phase, referred to as the NV Admission Control phase, can be based on regular call admission control schemes that can be found in current literature [21, 22]. In case of successful admission, the OpeNB notifies the MT-Host that it is capable of hosting the connections of the respective UE, by sending an ‘‘OpeNB Access Ack’’ message (step 10). Upon receiving the confirmation message from the OpeNB, the MT-Host initialize the procedure for the establishment of the L3 tunnel between (1) the VeNB instance at the target OpeNB and (2) a CN entity of Operator A. The S-GW Ctrl and P-GW-Ctrl controller applications running at the MC-Host send an appropriate OpenFlow (OF) forwarding rule to the S-GW and P-GW switches respectively, containing the instructions for forwarding the traffic relative to the tagged user towards the L3 tunnel to the home operator (steps 11–12). Once the tunnel endpoint at the host operator has been established, the MT-Host acknowledges the MT-Home confirming the admission of the tagged UE to the target OpeNB (step 13) providing also tunnel configuration information for the VeNB. Accordingly, the MC-Home configures the tunnel endpoint at the Home CN with the assistance of P-GW Ctrl and the S-GW Ctrl applications (steps 14 -15) and finally proceeds with the establishment of the L3 tunnel between the S-GW of the serving eNB and the target OpeNB, i.e. the respective VeNB instance. Note that the purpose of the L3 tunnel is to interconnect the VeNB at the host operator to the control plane and data plane entities at the CN of the home operator, i.e. the MME and S-GW, as it would have been a legacy eNB of the home operator. Once the tunnel establishment phase is completed (step 16), the VeNB forwards a NV-HO acknowledgement message towards the serving eNB through the L3 tunnel established with their common MME (steps 17 and 18). Upon reception of the NV-HO Acknowledgement, the serving eNB sends a HO command to the UE (step 19) and the UE performs legacy HO execution procedures towards the target OpeNB (step 20). Due to the presence of the VeNB instance at the OpeNB, the UE perceives the VeNB at the OpeNB as a legacy eNB that belongs to the home operator (operator A).
3.3 NV Phases In this section we provide a more detailed description of the NV phases mentioned in Sect. 3.2 (see the yellow boxes in Fig. 3).
3.3.1 NV Triggering Phase We start with the NV Triggering phase that is executed in step 3 (Fig. 3). Firstly, we note that HO triggering is integral part of the legacy HO execution procedure in LTE/LTE-A as
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
well, where the received signal strength at the UE, the received signal quality at the UE, or the cell load, are compared to absolute or relative thresholds. The key difference of the NV Triggering phase in our architecture, as compared to the legacy one, is that the set of candidate eNBs is enriched with the set of OpeNBs belonging to different operators. To this end, in this phase, both types of HOs are triggered: the ones towards a base station of the operator A (as in the legacy procedure) and the ones towards the OpeNBs of another operator. We consider that the choice of the NV triggering algorithm should be left open to the network operators, to enable them employ their own network management strategies. In the following, we propose two simple exemplary NV triggering algorithms: the NVRSRQ and the NV-Offloading algorithm. In the NV-RSRQ algorithm, the serving eNB triggers a NV-HO based on the Reference Signal Received Quality (RSRQ) measurements that are performed by the tagged UE and reported periodically in the serving eNB (steps 1–2). In more detail, the UE reports the RSRQ of all base stations in proximity. If the base station with the highest RSRQ belongs to the home operator, the serving eNB performs legacy HO execution. On the contrary, if the base station with the highest RSRQ belongs to another operator (OpeNB), the serving eNB sends a NV-HO request to the MC-Home and includes all reported RSRQ measurements by the UE. At this point, we note that this algorithm can be employed by reconfiguring the UE to perform RSRQ measurements in bands other than the ones of the home operator. This is possible in LTE/LTE-A since, on the one hand, the serving eNB indicates to the UE a list of frequencies for cell search [21] and, on the other hand, the UEs are capable of scanning and performing measurements in all bands available for cellular communications. In the NV-Offloading algorithm, the serving eNB triggers a NV-HO based on its cell load. In more detail, if the current cell load of the serving eNB is lower than a prescribed threshold, the serving eNB handles the execution of HOs as in the legacy HO execution phase. However, if the cell load of the serving eNB is higher than a threshold, the serving eNB triggers the UE to perform RSRQ measurements for all LTE-A base stations in proximity. Accordingly, the serving eNB follows a similar procedure with the NV-RSRQ algorithm. In more detail, if the base station with the highest RSRQ belongs to the home operator, the serving eNB performs legacy HO execution. Otherwise, it triggers a NV-HO towards the MC-Home. Let us now discuss the key advantages and weak aspects of both algorithms. The NVRSRQ algorithm is expected to achieve significant performance improvements at the UEs, since it enables the UEs to connect to the base station providing the most favorable signal quality conditions (regardless the operator to which it belongs to). At the same time, this can be the cause of increased leasing costs for the operator A as well as excessive traffic offloading towards the network of other operators. On the contrary, the NV-Offloading algorithm is expected to compensate these weaknesses as it performs a NV-HO only if the serving eNB is overloaded. Such an approach is expected to lower the number of NV-HOs in a network-wide scale and reduce the leasing costs of operator A.
3.3.2 NV Decision We now turn our attention to the NV decision phase that is performed in step 5 (Fig. 3). In this phase, the MC-Home decides on the most appropriate attachment point for the tagged UE in the cellular network of other operators, i.e. the S-RAN of the host operator in Fig. 3. This phase is performed only if the home operator has established SLA agreements with the host operator. To complete this phase, the serving eNB should provide the MC-Home
123
S. Costanzo et al.
with all information required to take the NV decision. Such information may include the characteristics of the ongoing UE connections, the list of reported measurements that triggered the NV- HO procedure, the load status at the serving eNB, the identity of the UE, and so on. In the remainder of this paper, we consider that the NV decision is taken based on the RSRQ status of the OpeNBs that are in proximity of the UE. In more detail, we consider that, upon NV-triggering, the MC-Home forwards a NV-HO request towards the OpeNB that belongs to the list of measurements provided by the serving eNB and attains the highest RSRQ for the tagged UE.
3.3.3 SLA Monitor In this phase, the MT-Host controller application checks if the NV-Request of the home operator is in line with the established SLA agreements. Let us consider an exemplary use case, where a third part infrastructure provider, i.e. the operator B, lets multiple network operators to lease its network infrastructure for the duration of a specific event, i.e. a sport event or a concert, in a specific geographic area. Under this model, one operator can improve its network capacity by having access to the additional resources from the base stations belonging to Operator B. In this scenario, the SLA Monitor should verify that the shared resources are allocated to each operator in a fair manner, without exceeding the limits specified in the SLA agreements. If the SLA monitor phase is successful, the MCHost forwards the NV-HO request of the home operator (operator A) to the target OpeNB.
3.3.4 NV Admission Control In this phase, the target OpeNB decides on whether to accept the NV-HO request, or not. The role of the NV Admission Control is to estimate the resource availability at the target OpeNB and infer on whether the target OpeNB can satisfy the quality of service (QoS) required by the tagged UE. In this phase, the OpeNB also estimates the impact of admitting a UE from another operator on the performance of the UEs that belong to host operator (Operator B). Note that the NV Admission Control phase can either give priority to the UEs of the host operator, or treat all attached UEs with the same priority, or perform cell-load based admission control.
3.4 Comments on the Signaling Load of the Proposed Multi-tenant Operation In this section we examine the signaling load of the proposed multi-tenant operation, i.e. the NV-HO from the home operator’s RAN (POP’s RAN) to the S-RAN of the host operator (MOP) and compare it with the legacy 3GPP LTE/LTE-A HO operation. In the legacy 3GPP LTE/LTE-A network the HO operation is driven by the serving eNB and most of the signaling messages to perform such an operation are exchanged between the serving eNB and the target eNB or between the eNBs and the MME [21]. Once the HO is executed all the data packets related to the user involved in the HO process are forwarded to the target eNB. Various requests and confirmation messages are required to be exchanged among the aforementioned entities. The number of these signaling messages increases for performing an inter-operator HO, e.g. the HO from the POP’s RAN to the S-RAN of the MOP. The employment of an SND approach can dramatically reduce the amount of signaling messages required for performing both intra-operator and inter-operator HOs. The SDN controller can leverage the benefit of the acquired global knowledge for lowering the signaling overhead required by the HO procedure. In fact the SDN
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
controller can take advantage of the centralized control of the SDN-based forwarding elements, i.e. the S-GW and P-GW in Fig. 1, for pushing forwarding rules at priori to all the entities involved in the HO operation reducing the need for them to exchange a multitude of request/ack messages as required by the legacy HO operation. Let us now focus on examine the NV- HO operation for the proposed framework. We consider an extension of the numerical analysis made in [17] for an SDN-based LTE-EPC so as to support the proposed NV-HO procedure. We focus on the signaling load at the control entities, i.e. the MME and the SDN-MC at the legacy 3GPP LTE/LTE-A architecture and at the enhanced architecture for RAN sharing respectively. According to [17] the signaling load is proportional to the number of signaling messages entering and leaving these control entities and the average arrival rate of user sessions. Looking to Fig. 3 it is clear that the total number of messages entering and leaving the SDN controller is 8. Conversely considering the HO procedure in the legacy 3GPP LTE-EPC, the signaling load at the MME can be calculated as for the following equation [17]: SLlegacy ¼
N X
20Rð1 Pi ÞB
ð1Þ
i¼1
where Pi is the probability for the user ‘‘i’’ to be in the idle state. B is the total number of eNBs in a region and R is the cross rate of the user i. Note that Eq. (1) is valid for the legacy HO operation with no X2 support [17, 21] that is the one similar to the proposed NV-HO, i.e. an S1-based handover. Conversely, the signaling load of the proposed NV-HO procedure (SLNV ) at the analog control plane entity, i.e. the SDN-MC, can be calculated as for the following equation: SLNV ¼
N X
8Rð1 Pi ÞB
ð2Þ
i¼1
Interestingly, looking to (1) and (2) it can be observed a decrease of the signaling load for the NV-HO procedure as compared to the legacy HO operation. This results from the employment of the SDN-based control plane that enables a considerable reduction of signaling messages compared to the ones required by the legacy 3GPP scenario. We also extend the numerical analysis presented in [17] for estimating the size of the SDN signaling messages for the proposed NV-HO procedure. Table 2 reports the size of the
Table 2 Signaling Messages size
SDN LTE-EPC Signaling message
Size (bytes)
4-Report UEs eNB-Ctrl
20
6-OpeNB HO request Ctrl–Ctrl
38
8-OpenB Access Request
62
10- OpeNB Access ACK
42
OF Tunnel FW rule (Ctrl-switch) ? (switch-Ctrl)
32 ? 24
NV-HO Ack OpeNB-POP_Ctrl
42
NV-HO Ack PoP_Ctrl- eNB
46
HO command
46
NV-HO confirmation
30
123
S. Costanzo et al.
signaling messages used in the proposed NV-HO procedure. In Sect. 4 we evaluate the signaling cost of the proposed NV-HO procedure by means of system level simulations.
4 Numerical Results In this section, we present system-level simulation results to assess the performance gains attained by the employment of the proposed framework in the LTE/LTE-A network. We consider a geographical area that includes two overlapped LTE-A networks owned by two different operators: Operator A and Operator B. The network of Operator A consists of 7 macro cells with inter-site distance of 1 km, while the network of Operator B consists of 7 macro cells that overlap the coverage of the macrocell network of Operator A. We further consider that Operator B owns a number of femtocells that are uniformly distributed within the coverage of the macrocells owned by Operator B. In more detail, we use the 3GPP 5X5 grid model [23] for the locations inside the buildings of a suburban area and assume that each building contains five femtocells that are uniformly distributed within the apartments of each building. The number of buildings is adapted depending on the desired femtocell density in the network and is used as the x-axis parameter in our simulation campaigns. In the sequel, we consider that Operator A acts as the home operator (POP), whereas the Operator B acts as the host operator (MOP). We also consider that all femtocell base stations of Operator B support the OpeNB functionality forming the so-called S-RAN (see Fig. 1) and operate in the same frequency of the Operator B’s macro cells. In the sequel we use the term ‘‘H-OpeNBs’’ to refer to the aforementioned femtocells. Note that the UEs Table 3 Simulation parameters Parameters
Values
Deployment #eNBs (Op.A)
7
Inter-site distance (Op.A)
1 km
eNB bandwidth
10 MHz (both Op. A and Op. B)
#macro eNBs (Op.B)
7
Inter-site distance (Op.B)
1 km
UEs density Operator A
40 UEs/Macro Cell
UEs density Operator B
30 UEs/Macro Cell
#Building per macro eNB (Op.B)
0 to 40
Building layout
3GPP 5X5 model
#Femtocells per building
5, uniformly distributed
#Femtocells in the system (H-OpeNBs)
0 to 1400
Femtocells bandwidth (Op.B)
10 MHz
Mobility model
Random Way Point Model [24]
UE speed
Pedestrian (3 km/h)
System model PATH loss model
L = 128.1 ? 37.6 log10(R) [24]
Fading model
Jakes model [24]
Scheduler downlink
MLWDF [24]
Application traffic model
Video Streaming (Download) encoded at 440 kbps [24]
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
registered to the Operator B can be served by either macro cells or femtocells of the Operator B, while the users registered to the Operator A are authorized to have access to the Operator B’s femtocells only after a successful NV-HO. We also assume that a fixed number of pedestrian UEs are uniformly distributed within the coverage area of each eNB operated by operator A (40 UEs/macrocell). We assume the network of the Operator B is less loaded (30 UEs/macrocell) so as to have enough resources available for the users handed over by the Operator A (NV-HO). The presented results are derived assuming that each UE sustains an H.264 video flow encoded at 440 kbps with a maximum delay constraint of 100 ms [24]. The simulations results were derived by using the LTE-Sim simulator [24]. The rest of the simulation parameters are summarized in Table 3 [23, 24]. We have extended a bit the architecture of LTE-Sim in order to enable the communication of the simulator architecture with the MC. More specifically, the MC consists of an external application that communicates with the simulator by means of sockets. Indeed the wired link between the controller and the simulator are simulated by means of an UDPbased socket channel. In this way we have emulated the southbound interface between the controller and the OpeNB. More specifically, the interaction between the protocol stack of the eNB and the MC is made by means of appropriate software agents, which have the role to translate the commands of the MC in 3GPP compatible messages for the L2/L3 layer of the eNB protocol stack. Note that in the following, the simulations focus only on the RAN, while not all the elements of the core network have been implemented, i.e. the simulator includes only the basic functionalities of the P-GW [24]. This because the performance evaluation of the core network is not under the focus of this paper, since we are more interested to evaluate the potential benefits of our solution at the RAN and user side. However, our future plan is to test our solution in a real testbed for 4G/5G, which includes also the core network, like OpenAirInterface [25]. In the sequel, we compare the performance of a legacy LTE network with that of an LTE network that employs the proposed framework. For brevity, we will refer to the first scenario as the baseline scenario, i.e. the one where the enhanced architecture is not employed. On the other hand, depending on the NV triggering and admission control algorithms adopted, we examine the performance of the proposed framework under three different NV scenarios. In the first scenario, which we tag as the ‘‘NV-RSRQ scenario’’, we employ the NV-RSRQ algorithm described in Sect. 3.3.1 for the NV triggering phase, and consider that the H-OpeNBs accept all the HO requests from Operator A in the NV Admission Control phase. In this scenario, all UEs are admitted as long as the target H-OpeNB has enough resources to satisfy their QoS requirements. In the second NV scenario, which we tag as the ‘‘NV-Offload’’ scenario, we employ the NV-Offloading algorithm described in Sect. 3.3.1 for the NV triggering phase and, once again, assume that the target H-OpeNBs accept all NV-HO requests from Operator A (as long as they can support the QoS requirements). In the third NV scenario, which we tag as the ‘‘NV-AC’’ scenario, we employ the NV-Offloading algorithm described in Sect. 3.3.1 for the NV triggering phase, and consider that the NV-HO requests from Operator A are accepted only if the traffic load at the H-OpeNBs is lower than a specific threshold., i.e. the 85% of the total PRB utilization in our simulations. We present different plots for the performance of the UEs that are registered in the home operator (Fig. 4) and the performance of the UEs that are registered in the host operator (Fig. 5). Nevertheless, in the first plot (Fig. 4) we also distinguish the performance of the UEs that are registered in the home operator and continue to receive service from their home operator, from the UEs that are registered in the home operator and exploit the proposed multi-tenant operation for receiving service from the host operator (Operator
123
S. Costanzo et al.
Fig. 4 Average DL Goodput for UEs registered to Operator A
Fig. 5 Average DL Goodput for UEs registered to Operator B
B). For brevity, we use the suffix ‘… : eNB users’ to refer to the first set of UEs and use the suffix ‘… : H-OpeNB users’ to refer to the second set of UEs. Figure 4 shows the average goodput of the UEs that are registered to the home operator, i.e. the throughput as measured at the application layer, for all scenarios under scope and for increasing number of H-OpeNBs (femtocells) in the host operator. As shown in Fig. 4, the UE goodput performance for all scenarios that employ the proposed framework, i.e. the ones with the prefix ‘NV-…’, outperform the UE goodput performance in the baseline scenario. This trend directly follows from the fact that the employment of the proposed framework enables the cellular UEs to connect to cellular BSs with more favorable channel conditions, e.g. higher RSRQ, or the lower path loss. Notably, we observe that the employment of the proposed framework, not only improves the goodput performance of the UEs that exploit the proposed NV-based solution, i.e. the ones offloaded to Operator B, but also improves the performance of the UEs that continue to receive service from Operator A. In fact, the performance of the UEs that continue to receive service from
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
Operator A is improved at a higher rate compared to the one of the UEs that are offloaded to Operator B. This follows from the fact that the proposed NV-based solution enables traffic offloading towards the host operator, leaving more resources for the UEs that continue to receive service from Operator A. Referring to the performance comparison between the different NV-scenarios under scope, we observe that the highest performance gains are achieved in the NV-RSRQ scenario (up to 14% gain) for both types of UEs registered to Operator A This behavior follows from the fact that the NV-RSRQ triggering algorithm favors the execution of HOs towards the base stations with more favorable channel conditions. Let us now focus on the performance of the NV-Offload and the NV-AC scenarios, in both of which we employ more restrictive policies during the NV Triggering and NV Admission Control phases. We first observe that the goodput performance of the UEs offloaded to Operator B is roughly the same for both the NV-Offload and the NV-AC scenario (up to 11% gain). On the contrary, a slightly better performance is observed for the UEs that continue to receive service from Operator B in the NV-Offload, as compared to the one of the NV-AC scenario. Interestingly, under low H-OpeNB densities (left side of the plot), the performance of the UEs offloaded to Operator B, under both the NV-Offload and the NV-AC scenarios, is higher compared to that of the UEs that continue to receive service from Operator A. However, this behavior alters in higher femtocell (H-OpeNB) densities, where an increased number of UEs from Operator A can be offloaded to the femtocells of Operator B, leaving more resources for the UEs that continue to receive service from Operator A. Appreciating the performance gains of the proposed framework for the UEs registered to Operator A (Fig. 4), let us now examine the average goodput at the UEs that are registered in Operator B (Fig. 5). As expected, the employment of the proposed framework reduces the average goodput at the UEs registered in Operator B, as a result of the increased demand of network resources, i.e. increased number of served UEs in Operator B. This observation readily follows by comparing the performance of the baseline scenario to that of the NV-based scenarios. Figure 5 also reveals that the highest gains at the UEs of Operator A (NV-RSRQ in Fig. 4) are attained at the cost of the highest performance losses (up to 5% throughput loss) at the UEs of Operator B (Fig. 5). Nevertheless, this mainly follows from the fact that the NV-RSRQ scenario assumes that the OpeNBs do not perform admission control at the Operator B. To this end, as shown Fig. 5, the employment of load-balancing based criteria during the NV triggering (NV-Offloading) and the NV admission control phases (NV-AC) can result in notable performance gains for the UEs registered to Operator A (Fig. 4) without significantly deteriorating the performance of the UEs registered to Operator B (only 1% goodput loss for the NV-AC scenario). In Fig. 6, we plot the average DL SINR for the UEs registered to the home operator (averaged over all users). As the number of H-OpeNBs increases, an enhanced DL SINR is experienced at the UEs registered to Operator A, including both the ones that continue to receive service from Operator B and the ones that are offloaded to Operator B. This performance improvement follows from the flexibility offered to the UEs registered to Operator A, to associate with base stations owned by different mobile network operators. As in Fig. 4, the highest performance gains (up to 12% gain) are shown to be attained for the NV-RSRQ scenario (close to 1.5 dB), whereas notable performance gains are observed for the NV-Offloading and NV-AC scenarios as well (close to 1.2 and 1 dBs, respectively). At this point, it is important to note that in our simulations we have assumed that the macrocell and the femtocell stations (H-OpeNBs) belonging to Operator B, operate in the
123
S. Costanzo et al.
Fig. 6 Average DL SINR for UEs registered to Operator A
same frequency. Hence, even higher performance gains would be expected if the femtocell and macrocell base stations in Operator B utilized different frequencies. In Fig. 7, we plot the average DL SINR at the UEs registered to the host operator, for the different scenarios under scope and for increasing H-OpeNB density per macrocell. As expected, the admission of additional UEs in Operator B, i.e. the ones utilizing the proposed multi-tenant operation, reduces the average DL SINR for all UEs registered to Operator B. Nevertheless, as compared to the performance gains attained for the UEs registered at the home operator (Operator A), the performance loss is comparably lower (up to 3% loss), and strongly depends on the NV-scenario under scope. In more detail, the NV-RSRQ scenario is shown to reduce the average DL SINR at the UE registered in Operator B by up to 0.5 dB. On the contrary, both the NV-Offloading and the NV-AC scenarios are shown to attain slightly lower performance as compared to the baseline scenario, where the NV-AC can be said to attain roughly the same performance for all densities under scope.
Fig. 7 Average DL SINR for UEs registered to Operator B
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
Fig. 8 Average packet delay for UEs registered to Operator A
Figure 8 depicts the average end-to-end delay experienced by the UEs registered to the home operator (Operator A), as measured at the application layer. We firstly observe that the employment of the proposed framework significantly reduces the end-to-end application-layer delay at the UEs registered to Operator A for all NV scenarios under scope. Secondly, we observe that the performance gains are proportional to the number of femtocells (H-OpeNBs) available from the host operator (Operator B). Thirdly, Fig. 8 reveals that the end-to-end application-layer delay at the UEs is roughly the same for all the NV scenarios under scope when the H-OpeNB density at the host operator is medium to high, i.e. higher than 160 H-OpeNBs per macrocell. Another interesting observation is that apart from increased goodput at the application layer (Fig. 5), the employment of the proposed framework results in substantial reduction of the end-to-end application-layer delay at the UEs (Fig. 7). This reduction reaches up to 57% as compared to the baseline scenario. Such performance gains can significantly enhance the experience of the end-user, upon reception of delay-sensitive services, fully capitalizing the performance gains following from shifting NV to the access network of the LTE-A system. It should be noted that the performance improvement, in terms of end-to-end application-layer delay, not only follow from the enhanced goodput attained at the UEs (Fig. 4), but also from the flexibility that enables the UEs to associate with the closest base station in proximity (even if it belongs to a different operator). Besides, in lower loads, the packet scheduler at the BSs of the home operator can better handle the packet flows of the UEs that remain in the home operator. This effect further reduces the queue processing time, decreasing the overall packet delivery latency as well. Let us now examine the impact of offloading a certain number of UEs to the host operator, in terms of end-to-end application layer delay at the UEs registered to the host operator (Fig. 9). As expected, the end-to-end application-layer delay at the UEs registered to Operator B is reduced for higher H-OpeNB deployment densities (baseline scenario). This behavior is common for all scenarios under scope, since the deployment of additional femtocell stations at the host operator, increases the network density and reduces the mean distance to the nearest base stations for all UEs served by the network of Operator B. Interestingly, even though the employment of the proposed NV-base architecture is shown to increase the average end-to-end application-layer delay for the UEs registered to Operator B, as the H-OpeNB deployment density increases, this performance deterioration
123
S. Costanzo et al.
Fig. 9 Average packet delay: UEs registered to Operator B
is comparably smaller than the performance gains attained at the UEs registered to Operator A (Fig. 8). The employment of the NV-RSRQ scenario, which has been shown to provide up to 21 ms reduced delay for the UEs registered to Operator A, is shown to increase the end-to-end delay at the UEs registered to Operator B by up to 9 ms. On the other hand, the performance of the NV-AC and the NV-Offloading scenarios is shown to be roughly the same as compared to the baseline scenario (up to 2 ms increase of the endto-end delay). At this point, it is important to note that the performance deterioration at the UEs of Operator B will leave their QoS performance unaffected, since the maximum delay requirement for the assumed traffic type, i.e. delay-demanding video streaming, is 100 ms. In Fig. 10, we plot the average network load at the home and the host operator that results from the traffic offload operation provided by the proposed NV-HO procedure. Let us consider an exemplary use case scenario, such as a peak demand event (e.g. a sport event or traffic jam) where a network operator, i.e. Operator A, needs to increase its network capacity to satisfy the increasing traffic demands within small geographical areas. Let us also assume that Operator A (home operator) has to serve a fixed number of
Fig. 10 Average network load
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
pedestrian users, i.e. 280 UEs, for the duration of such event. In this specific scenario, the home operator may benefit of the proposed framework since it enables the dynamic increase of its network capacity by offloading traffic to the base stations of another operator, i.e. Operator B (host operator). The amount of traffic that the operator A may offload to the host operator depends of the specific NV policy that is established in the predefined SLA agreement. Let us now analyze the results in terms of network load that derive from employing each NV scenario under scope. Note that in the sequel we assume that, initially, the Operator B has to serve a lower number of UEs (210) as compared to the baseline scenario of the Operator A. As anticipated in the previous results, the NV-RSRQ scenario allows the home operator to offload a higher number of UEs to the host operator, i.e. up to 25% of the served UEs. This is due to the fact that the NV triggering algorithm employed at this NV scenario favors a higher number of NV-HO towards the host operator. However, this results in an up to 34% increase of the traffic load at the host operator. Besides, in the other two NV scenarios we observe a lower percentage of traffic offload, i.e. up to 20% decrease at the home operator and up to 28% increase at the host operator. We also note that in the NV-Offload and NV-AC scenario we can achieve almost the same results for low femtocell densities, while a rough difference is observed for higher densities. In fact, these two scenarios employ the same NV triggering algorithms that results in the same number of triggered NV-HOs. However the NV-AC admits a lower number of UEs at the NV-AC phase, especially in high femtocell density scenarios where the NV-HO probability increases. We now focus on the signaling cost required to support the proposed NV-HO procedure. We compare the costs in the proposed NV-HO procedure for the home operator with the home operator’s baseline scenario, i.e. the single home operator network scenario with no access to the multi-tenant base stations of the host operator. Moreover we evaluate a third scenario, referred to as Legacy-MT, which consists of a heterogeneous network of macrocells and femtcocells. The macro cell network corresponds to the home operator network in the baseline scenario, while the network of femtocells corresponds with the one deployed by the host operator in the S-RAN with the difference that all femtocells are supposed to be owned by the home operator and the legacy 3GPP LTE/LTE-A architecture is employed instead of the proposed framework. This means that all the HOs in LegacyMT, including the ones toward the femtocells, are managed in a legacy 3GPP LTE/LTE-A fashion. For the Legacy-MT scenario we evaluate the HO signaling costs as for a legacy 3GPP architecture given the HO rate resulting from employing one the aforementioned NV scenario (NV-AC). The metric we use for evaluating the signaling costs is: Pn signal pkti pkt size ð3Þ sign cost ¼ i¼0 DT where i is the i-th signaling packet, pkt_size is the size of the i-th signaling packet and DT is the time interval. We refer to Table 2 for the size of the signaling messages of the proposed NV-HO procedure, while we refer to the legacy 3GPP HO procedure [21] for the size of the signaling messages in both baseline and Legacy-MT scenarios. From Fig. 3 it can be observed that 20 signaling messages are required for the first HO from the POP’s RAN to the S-RAN. Note that from the second HO to on, the signaling messages required for the tunnel establishment phase are not required anymore because the forwarding table at the S-GW and PG-GW in both MOP and POP CNs have already been configured with an appropriate rule for handling the user traffic related to the POP’s users.
123
S. Costanzo et al.
Figure 11 shows the HO signaling cost for home operator in the baseline, Legacy-MT and NV-AC scenario respectively. As expected in the baseline scenario the HO signaling cost is lower than in the NV-AC. This result comes from the lower base station density in the baseline scenario that leads to a less number of HOs as compared to the NV-AC. In fact the signaling cost is proportionally to the HO rate, which is in general higher in a denser network scenario like the proposed multi-tenant scenario. However, despite the higher HO rate, in the NV-AC scenario the signaling cost results only 5% higher than the baseline for lower H-OpeNB density. Moreover it can be observed the NV-AC scenario shows a considerable performance gain as compared to the Legacy-MT scenario. Accordingly, we note that the use of the SDN-based LTE-EPC requires up to 17% less signaling cost as compared to the legacy 3GPP scenario given the same network densification scenario (Legacy-MT). This results from the lower number of signals that are required by the SDN control plane as discussed in Sect. 3.4. Even though this numeric evaluation is a simple attempt to assess the potential benefits of SDN in LTE networks, it encourages the choice of SDN as a key element for building next generation mobile network architecture as it brings not only more flexibility in the control plane management but it also leads to a notable improvement of the network performances, e.g. signaling load, throughput, delays, etc. Let us now focus on evaluating the potential profit achievable by both the operators involved in multi-tenant scenario depicted in Fig. 2: the host operator (MOP) that puts at disposal a network of H-OpeNBs (S-RAN) to one or more participating operators and the home operator (POP) that leases the resources of the S-RAN on demand. For the sake of simplicity we consider the system (aggregate) throughput as a metric to evaluate the operator’s revenue. In fact a higher system throughput usually produces a higher income for the operator, e.g. because more users can be served at the same time and/or better QoS/ QoE can be offered to a set of premium users that are willing to pay more for getting a better service. In the sequel we compare the system throughput for both the POP and MOP operators in the baseline and NV-AC scenarios respectively. From the business model perspective, we assume that MOP charges the POP with a leasing fee that is proportionally to a percentage of the offered throughput to the POP’s users served by the S-RAN. Accordingly, the POP’s revenue (RPOP ) takes into account the
Fig. 11 Handover signaling cost
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations
throughput of all the users registered to the POP (TPOP ) plus a percentage of throughput (L) of the users served by the S-RAN (TSRAN ) as result of a NV-HO, as per Eq. (4): POP SRAN RPOP ¼ TPOP L TPOP
½X
ð4Þ
where X is the revenue unit per Mb. Accordingly the revenue for the MOP (RMOP ) takes into account the throughput of all the users registered to the MOP (TMOP ) plus a percentage of the throughput of the users that are handed over from the POP’s RAN to S-RAN, as per Eq. (5). SRAN RMOP ¼ TMOP þ L TPOP
½X
ð5Þ
For the sake of simplicity in our simulations we assumed L = 10% in Eqs. (4)–(5), however such a parameter can be easily changed and adapted for a multitude of business models scenarios. Figure 12 shows the RPOP for the home operator (Operator A) in the baseline and NV-AC scenario respectively. As it can be seen from Fig. 12, the RPOP proportionally increases with the H-OpeNB density and an up to 13% gain is observed for the NV-AC scenario as compared to the baseline scenario. This result demonstrates the profitability of the proposed NV-based solution for the POP operator. In fact, although the POP has to pay a leasing fee for using the S-RAN, a notable improvement of the system throughput can be achieved by offloading user traffic to the S-RAN. This also enables the POP to face the increasing demand for traffic in a cost-effective manner. Figure 12 also reveals the RMOP for the host operator. When no femtocells (H-OpeNBs) are deployed the RMOP comes only from the non-shared RAN consisting of a macro cell network that serves only the users registered to host operator. By increasing the density of the H-OpeNBs, the RMOP increases proportionally to the number of users that are handed over from both the POP’s RAN and MOP’s non-shared RAN. In fact, the deployment of the femtocell network lets the MOP to increase the system throughput by offloading traffic from the macrocell network (non-shared MOP’s RAN) and gets also a profit from leasing the resources of these femotcells (H-OpeNBs) to the POP. The total operator revenue, calculated according to (5), finally results higher in the NV-AC than in the baseline scenario (where the H-OpeNBs are exclusive used by the MOP’s users), even with the
Fig. 12 Revenue for POP and MOP
123
S. Costanzo et al.
simple business model we consider in this paper, wherein the POP pays a leasing fee to the MOP that is proportional to the TSRAN . To conclude our analysis, we note that even POP though the proposed NV-based solution may result in a slight deterioration of the performances at the UEs registered to the MOP operator (see Fig. 5), the profit that can be achieved from the sharing process is expected to be much higher for both MOP and POP operators.
5 Conclusion In this paper, we have proposed a framework that enables multi-tenancy in evolving LTEA networks by employing SDN/NFV technology. The proposed framework launches the design of innovative business models enabling the mobile operators to create an augmented RAN on demand by utilizing the base station infrastructure of other operators. Moreover we have proposed a NV-based solution for multi-tenancy that enables operators to bring the access network closer to their UEs, while it can also be used to offload the traffic to the base stations of another network operator (host operator) under high duty cycles, e.g. crowded public events. The proposed NV-based solution is accompanied by a comprehensive discussion on the architectural and functional enhancements required to the existing LTE/LTE-A RAN architecture. A detailed specification of the signaling procedure has also been presented for supporting the proposed framework in a real-life network. Extensive system-level simulations have been used to assess the performance of the proposed framework and compare it to a legacy LTE/LTE-A network. In more detail, the simulation results have shown that, even under the employment of simple NV-triggering algorithms, the proposed framework can significantly improve the goodput and end-to-end delay performance at the UEs registered to the home operator. Nevertheless, these performance gains are attained at the cost of a slight increase in signaling rate between the two operators and small performance deterioration at the UEs registered to the host operator. Notably, simulation results have also shown that the performance of the UEs registered to the host operator can remain roughly unaffected if simple admission control schemes are employed at the base stations that are endowed with multi-tenant capabilities. In our future work, we aim to propose more effective NV-triggering and resourcesharing strategies among the operators involved in the proposed framework. Within our scope, will be also to shed more light on how the proposed NV-based solution can be exploited in the context of evolving the LTE-A network architecture towards the 5G RAN architecture. Moreover, our future plan is to test our solution in a real testbed for 4G/5G, which includes also the core network, like OpenAirInterface [25]. We also aim at validating the OpeNB solution with a real SDN controller, like the one that is going to be developed in the context of the FlexRAN project [26] and investigate in more details the benefits of our solution taking into account the network elements of the core network. Acknowledgements This paper has been funded by the FP7 Marie Curie project CROSSFIRE (MITN 317126).
References 1. Gupta, A., & Jha, R. K. (2015). A survey of 5G network: Architecture and emerging technologies. Access, IEEE (vol. 3, pp. 1206–1232).
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations 2. Bhushan, N., Li, J., Malladi, D., Gilmore, R., Brenner, D., Damnjanovic, A., et al. (2014). Network densification: The dominant theme for wireless evolution into 5G. Communications Magazine, IEEE, 52(2), 82–89. 3. GPP TS 23.251, ‘‘Network sharing; Architecture and functional description’’, Rel. 13, Mar. 2015. 4. GPP TS 32.130, ‘‘Telecommunication management; Network sharing; Concepts and requirements’’, Rel. 12, Dec. 2014. 5. Chowdhury, N. M. M. K., & Boutaba, R. (2009). Network virtualization: State of the art and research challenges. Communications Magazine, IEEE, 47(7), 20–26. 6. Costa-Perez, X., Swetina, J., Tao, G., Mahindra, R., & Rangarajan, S. (2013). Radio access network virtualization for future mobile carrier networks. Communications Magazine, IEEE, 51(7), 27–35. 7. O. M. E. Committee, ‘‘Software-defined Networking: The New Norm for Networks’’, Open Networking Foundation, 2012. 8. ‘‘Network Functions Virtualisation-Introductory White Paper’’, ETSI, October 2012. 9. Panchal, J. S., Yates, R. D., & Buddhikot, M. M. (2013). Mobile network resource sharing options: Performance comparisons. Transactions on Wireless Communications, IEEE, 12(9), 4470–4482. 10. Kiess, W., Weitkemper, P., & Khan, A. (2013). Base station virtualization for OFDM air interfaces with strict isolation. In IEEE International Conference on Communications (ICC), Budapest. 11. Zaki, Y., Zhao, L., Go¨rg, C., & Timm-Giel, A. (2011). Lte mobile network virtualization. Mobile Networks and Applications, 16(4), 424–432. 12. GPP - 5G Vision Brochure, available on-line at www.5g-ppp.eu. 13. Costanzo, S., Xenakis, D., Passas, N., & Merakos, L. (2014). OpeNB: A framework for virtualizing base stations in LTE networks. In IEEE International Conference on Communications (ICC), Sydney. 14. Shrivastava, R., Costanzo, S., Samdanis, K., Xenakis, D., Grace, D., & Merakos, L. (2014) An SDNbased framework for elastic resource sharing in integrated FDD/TDD LTE-A HetNets. In IEEE international conference on cloud networking (CloudNet), Luxembourg. 15. Bradai, A., Singh, K., Ahmed, T., & Rasheed, T. (2015). Cellular software defined networking: A framework. Communications Magazine, IEEE, 53(6), 36–43. 16. Sama, M. R., Contreras, L. M., Kaippallimalil, J., Akiyoshi, I., Haiyang, Q., & Hui, N. (2015). Softwaredefined control of the virtualized mobile packet core. Communications Magazine, IEEE, 53(2), 107–115. 17. Nguyen, V. G., & Kim, Y. H. (2015). Proposal and evaluation of SDN based mobile packet core network. Journal on Wireless Communications and Networking, EURASIP, 1, 1–18. 18. McKeown, N., Anderson, T., Balakrishnan, H., Parulkar, G., Peterson, L., Rexford, J., Shenker, S., & Turner, J. (2008). OpenFlow: Enabling innovation in campus networks. In Computer communication review, ACM SIGCOMM (vol. 38, pp. 2). 19. China Mobile Research Institute. (2011). C-RAN: The Road Towards Green RAN, White Paper, Version 2.5. 20. Gudipati, A., Perry, D., Li, L. E., & Katti, S. (2013). SoftRAN: Software defined radio access network, HotSDN, ACM SIGCOMM. 21. Xenakis, D., Passas, N., Merakos, L., & Verikoukis, C. (2014). Mobility management for Femtocells in LTE-advanced: Key aspects and survey of handover decision algorithms. In Communications Surveys and Tutorial, IEEE (vol. 16(1), pp. 64–91) First Quarter 2014. 22. Kuklinski, S., Yuhon, L., & Khoa, T. D. (2014) Handover management in SDN-based mobile networks. In IEEE Globecom Workshops (GC Wkshps), Austin. 23. GPP TS 36.814, ‘‘Evolved Universal Terrestrial Radio Access (E-UTRA); Further advancements for E-UTRA physical layer aspects’’, Rel. 9, Mar. 2010. 24. Piro, G., Grieco, L. A., Boggia, G., Capozzi, F., & Camarda, P. (2011). Simulating LTE cellular systems: An open source framework. In Transactions on Vehicular Technology, IEEE (Vol. 60, pp. 2). 25. OpenAirInterface. http://www.openairinterface.org. 26. Foukas, X., Nikaein, N., Kassem, M. M., Marina, M. K., & Kontovasilis, K. (2016). FlexRAN: A flexible and programmable platform for software-defined radio access networks. In ACM Conference on emerging Networking EXperiments and Technologies (CoNEXT), Dec. 2016.
123
S. Costanzo et al. Salvatore Costanzo received his B.Sc. in ‘‘Telematics Engineering’’ in 2007 and his M.Sc. in ‘‘Telecommunications Engineering’’ in March 2012, both from the University of Catania, Italy. From 2009 to 2013 he was a research engineer at the Italian National Inter-University Consortium for Telecommunications (CNIT). With CNIT he was involved in various Italian and European research project, including ‘‘CRUISE’’, ‘‘NEWCOM??’’ and ‘‘SENTINELLA’’. Since 2013 he is Early Stage Researcher in the context of FP7 Marie Curie project CROSSFIRE and a PhD candidate in the Department of Informatics and Telecommunications of the University of Athens. His research interests focus on analysis and solutions for ad-hoc and mobile networks, with a particular emphasis on network virtualization and software-defined network architectures.
Dionysis Xenakis received his B.Sc. degree in Computer Science in 2007, his M.Sc. degree in Communications Systems and Networks in 2009, while he has pursued his Ph.D. degree at the Department of Informatics and Telecommunications, University of Athens, Greece. In 2008, he received the M.Sc. Excellence Award in the field of Networks and Communication Systems from the same department. Dionysis participated in various FP7 research projects, including PHYDYAS, C2POWER, HRAKLEITOS II, CROSSFIRE, and SMART-NRG. He is a co-author of 9 conference papers, 5 journals papers, and 4 book chapters, while he has also been a reviewer in numerous peer-reviewed conferences (e.g., IEEE GLOBECOM, VTC, ICC, and WCNC), and journals (e.g. IEEE TWC, IEEE TCOM, IEEE TVT, IEEE ComMag, IEEE WCM, IEEE CL, Elsevier ComCom, Elsevier ComNet, Springer WINET). He has given numerous seminars for training under various EU-funded projects, such as BeFEMTO, FREEDOM, CROSSFIRE, ACROPOLIS, WHERE2 and others, while he has also served as a TPC member in various IEEE conferences such as IEEE ICC, IEEE CAMAD, IEEE ISWTA, IEEE ISCI and IEEE HealthCom. His current research interests include Mobility Management for Femtocells Networks, Localization in Heterogeneous Networks, and D2D Communications. Dionysis is currently an IEEE student member, member of the Communication Network Laboratory/University of Athens, Greece, and member of the Green Adaptive and Intelligent Networking Group/University of Athens, Greece. Nikos Passas received his Diploma (honors) from the Department of Computer Engineering, University of Patras, Greece, and his Ph.D. degree from the Department of Informatics and Telecommunications, University of Athens, Greece, in 1992 and 1997, respectively. He is currently a member of the laboratory teaching staff in the Department of Informatics and Telecommunications of the University of Athens, and a group leader of the Green, Adaptive and Intelligent Networking (GAIN) research group inside the department. Over the years, he has participated and coordinated a large number of national and European research projects. Dr. Passas has served as a guest editor and technical program committee member in prestigious magazines and conferences, such as IEEE Wireless Communications Magazine, Wireless Communications and Mobile Computing Journal, IEEE Vehicular Technology Conference, IEEE PIMRC, IEEE Globecom, etc. He has published more than 120 papers in peer-reviewed journals and international conferences and has also published 1 book and 11 book chapters. His research interests are in the area of mobile network architectures and protocols. He is particularly interested in quality of service provision for wireless networks, medium access control, and mobility management. Dr. Passas is a member of the IEEE and a member of the Technical Chamber of Greece.
123
Augmented RAN with SDN Orchestration of Multi-tenant Base Stations Lazaros Merakos received the Diploma in Electrical and Mechanical Engineering from the National Technical University of Athens, Athens, Greece, in 1978, and the M.S. and Ph.D. degrees in Electrical Engineering from the State University of New York, Buffalo, in 1981 and 1984, respectively. From 1983 to 1986, he was on the faculty of the Electrical Engineering and Computer Science Department, University of Connecticut, Storrs. From 1986 to 1994, he was on the faculty of the Electrical and Computer Engineering Department, Northeastern University, Boston, MA. During the period 1993–1994, he served as Director of the Communications and Digital Processing Research Center, Northeastern University. During the summers of 1990 and 1991, he was a Visiting Scientist at the IBM T. J. Watson Research Center, Yorktown Heights, NY. In 1994, he joined the faculty of the University of Athens, Athens, Greece, where he is presently a Professor in the Department of Informatics and Telecommunications, and Scientific Director of the Networks Operations and Management Center. His research interests are in the design and performance analysis of communication networks, and wireless/mobile communication systems and services. He has authored more than 200 papers in the above areas. He has served as the scientific director of the Communication Networks Laboratory of the University of Athens in numerous research projects, including the projects RAINBOW, WAND, MOBIVAS, WINE, EURO-CITI, POLOS, ANWIRE, E2R, E2RII, E3, Self-NET funded by the European Union. Dr. Merakos is chairman of the board of the Greek Schools Network, and member of the board of the National Research Network of Greece. In 1994, he received the Guanella Award for the Best Paper presented at the International Zurich Seminar on Mobile Communications.
123
