See discussions, stats, and author profiles for this publication at: [Link]

net/publication/366815511

On The Benefits of vDU Standardization in Softwarized NG-RAN: Enabling

Technologies, Challenges, and Opportunities

Article in IEEE Communications Magazine · April 2023

DOI: 10.1109/MCOM.001.2200390

CITATIONS READS

12 337

3 authors:

Behnam Ojaghi Ferran Adelantado Freixer

Polytechnic University of Catalonia Universitat Oberta de Catalunya
17 PUBLICATIONS 173 CITATIONS 80 PUBLICATIONS 2,588 CITATIONS

SEE PROFILE SEE PROFILE

Christos Verikoukis
University of Patras
551 PUBLICATIONS 9,409 CITATIONS

SEE PROFILE

All content following this page was uploaded by Behnam Ojaghi on 05 January 2023.

The user has requested enhancement of the downloaded file.

 1

On The Benefits of vDU Standardization in

Softwarized NG-RAN: Enabling Technologies,
Challenges, and Opportunities
Behnam Ojaghi, Ferran Adelantado, and Christos Verikoukis

Abstract—Softwarization of Next Generation-Radio Access as network slice instantiating and maintenance, slicing over
Network (NG-RAN) comes along with virtualization of Network multi-domains, allocation of computing, storage, and radio
Functions (NFs) as purely software-based applications, dis- resources along with supporting algorithms and mechanisms.
tributed among a Central Unit (CU) and Distributed Units (DU).
This paper demonstrates the benefits of NG-RAN softwarization Furthermore, it is anticipated that NG-RAN architecture will
that enable Mobile Network Operators (MNOs) to adopt cutting- support a variety of services with high flexibility, satisfying
edge technologies and maximize network efficiency. In this article, various needs including latency, data rates, and extensive
we focus on the benefits of standardization of virtualized Dis- connectivity. In that sense, a NF offers a specific capability to
tributed Unit (vDU), and describe NG-RAN features intended to support these services. NFs can be fundamental operations that
boost network performance and maximize resource efficiencies in
deployment, operation, and management, resulting in a reduced are common to all applications, such as identity management
Total Cost of Ownership (TCO), particularly when it comes to and authentication NFs. On the other hand, some additional
addressing different needs of new services in NG. NFs may not apply to all use-cases. For example, a mobility
management NF like handover is only utilized for eMBB ser-
vices, or a user of an uRLLC requires more decentralized NFs
I. I NTRODUCTION to minimize the Hybrid Automatic Repeat Request (HARQ)
The Fifth-Generation (5G) of mobile communications was delay and ensure low delay [1].
promoted as an enabler for the Enhanced Mobile Broad- The concept of network softwarization will be a key tech-
band (eMBB) services with high-bandwidth requirements e.g., nology for NG-RAN. Leveraging the advantages of SDN/NFV
Virtual Reality (VR), ultra-Reliable and Low-Latency Com- in the radio access domain, which offers significant cost re-
munications (uRLLC) services demanding low-latency and ductions, high flexibility, and openness to promote innovation,
high-reliability e.g., Vehicle-to-everything (V2X), and massive and is anticipated to serve diverse services and sectors in NG.
Machine Type Communications (mMTC) services for verti- However, it poses several challenges that are being investigated
cal industries such as a smart city. eMBB services, among by numerous organizations, including the O-RAN Alliance,
other promises, have gained substantial interest in several Rakuten’s greenfield deployment in Japan, Small Cell Forum
5G deployments, like employing millimeter Wave (mmWave) (SCF)’s and Cisco’s Open RAN Platform, Samsung’s vRAN
frequency band in the Next Generation-Radio Access Network ecosystem, and Facebook Telecom Infra Project’s vRAN fron-
(NG-RAN). Initial deployments of the uRLLC and mMTC thaul Project Group. Arguably, among these initiatives, O-
services indicate that the 5G system supports their fundamental RAN has the highest popularity.
functionalities; nevertheless, support for complex use-cases Despite the lack of definition of new technologies and
of such services remains fairly challenging. These shortages requirements accepted by the standardization community, Aca-
stem mainly from the existing limits in the adaption of demic and industrial researchers are focusing on different
latest technologies in NG-RAN. Moving towards the NG characteristics of NG networks. One of the critical issues,
era, a wide variety of cutting-edge technologies and services among others, is defining the requirements and key enabling
will be required to adapt to the new architecture such as technologies to be standardized in NG-RAN architecture.
New Radio-Dual Connectivity (NR-DC), Network Function With this respect, we intend to (i) deliver an overview of
Virtualization (NFV), Software-Defined Networking (SDN), standardization of NG-RAN architecture, the motivation for
containerization, cloudification, softwarization, Multi-access their redesign pertaining to NG-RAN requirements; (ii) present
Edge Computing (MEC), network slicing just to name a few. an overview of recent technologies that could be well-adopted
These technologies are still in their infancy to be integrated into NG-RAN architecture and creation of service-specific
into NG-RAN architecture. In particular, MEC, among key RAN slices; (iii) outline the major drivers for mMTC, uRLLC,
enablers for NG-RAN, offers a computational capacity of and eMBB services; and (iv) identify key challenges and
decreased latency at the edge to support end-users. Network benefits of vDU standardization.
slicing is another important component of NG-RAN architec-
ture, enabling Mobile Network Operators (MNOs) to create II. T OWARDS S TANDARDIZATION OF NG-RAN
virtualized networks which could be customized to fulfill a A RCHITECTURE
diverse range of requirements in terms of functionality and From a network architecture perspective, the protocol stack of
isolation. They introduce significant technical challenges such RAN node is divided into several layers, each of which is in
 2

charge of a particular NF or a set of NFs. This section first Protocol (PDCP) layer. The northbound interface is backhaul
presents the RAN architecture evolution and then discusses network to the core network; and the southbound interface
the RAN splits standards for NG-RAN. is F1, that sits between O-CU and O-DU and is referred to
as midhaul as defined by 3rd Generation Partnership Project
A. Evolving RAN Architecture (3GPP).
The RAN has evolved over Distributed RAN (D-RAN)
deployments with co-located radios, i.e., Remote Radio 2. O-RAN Distributed Unit (O-DU): is responsible for all
Head (RRH), and BaseBand Units (BBU). These BBUs BBU processing, Radio Link Control (RLC), Medium Access
are connected via backhaul networks to the core. In 5G, Control (MAC), and the upper part of Physical (PHY) layer.
RAN deployment architectures evolved into i) Cloud RAN The northbound interface is F1, and the fronthaul network
(C-RAN), ii) virtualized RAN (vRAN), and finally, Open is southbound interface. Virtualizing this component asks for
RAN (O-RAN). extra hardware assistance in the form of accelerators such
as Field-Programmable Gate Arrays (FPGAs) or Graphics
C-RAN: firstly proposed by China Mobile [2] and Processing Units (GPUs). To enable full decoupling of the
decomposed into three major units: CUs with highly O-CU and O-DU software from the acceleration hardware,
programmable processors; RRHs and fronthaul network, Acceleration Abstraction Layer-Application Programming In-
connecting CU and RRHs. In this network architecture, Base terfaces ((AAL-API) hides hardware differences and provides
Station (BS) is entirely migrated from RRHs into a centralized a unified API to O-CU and O-DU software.
CU. A fronthaul network connects all RRHs to CUs, while a 3. O-RAN Radio Unit (O-RU): is in charge of the lower
backhaul network connects CUs to the Evolved Packet Core part of PHY layer such as Fast Fourier Transform (FFT)
(EPC). CUs transmit radio signals through Common Public and Inverse Fast Fourier Transform (IFFT). O-RU might
Radio Interface (CPRI) or Open Base Station Architecture potentially be virtualized, but the O-RAN Alliance working
Initiative (OBSAI). group is planning to use COTS components for a ”white box”
radio implementation.
vRAN: incorporates SDN/NFV virtualization techniques
into the RAN, hence, virtualizing all network functions (i.e., In O-RAN architecture, Service Management and
Virtualized Network Functions (VNF)) and resources along Orchestration (SMO) framework comprises the Non-Real
with decoupling Data Plane (DP) and Control Plane (CP) Time RAN Intelligent Controller (Non-RT RIC) function,
among CU and DU. The DU places near RRHs and facilitates which aims to support intelligent RAN optimization in
real-time processing, such as communication between the non-real-time (i.e., greater than one second) with the help of
user and cell site. It is, however, still in its early stages of data analytics, Artificial Intelligence, and Machine Learning.
development. Virtualization of the spectrum, air-interface Near-RT RIC supports near-real-time optimization of O-
and infrastructure, multi-Radio Access Technologies (m- RAN resources (O-CU and O-DU) over E2 interface with
RATs) virtualization, and computing resources are only a near-real-time control loops (i.e., 10ms to 1s) and hosts
few examples of various levels of virtualization that are xApps to obtain near-real-time RAN information to deliver
feasible. For example, BBU processing can be virtualized and customized services and data enhancements given by A1
disaggregated among CU and DU elements, allowing to run interface from Non-RT RIC. This architecture helps MNOs
VNFs to be implemented on standardized, Commercial-Off- in assisting the diverse needs in 5G networks by equipping
The-Shelf (COTS) hardware like x86 or Arm-based servers. the customized Functional Split (FS) per use case due to
Hence, vRAN, through the use of commodity hardware, open and interoperable interfaces, which adds benefits to both
offers more flexibility and reduces RAN costs. network systems and user experience.

O-RAN: is dedicated to promoting RAN which dismantles NG-RAN: is the latest RAN architecture indicated by
what used to be a one-size-fits-all, hardware-centric RAN, 3GPP in Release 15 combined with 5G mobile network and
delivering more open, interoperable interfaces and elastic than its latest standards were made available in 3GPP Release 16
existing deployed networks. In that sense, O-RAN Alliance [4], which details new interfaces, FS options, terminologies,
is working toward NG networks, where MNOs employ functional components, etc. in beyond 5G network. We will
standardized interfaces to control multi-vendor infrastructures further discuss the FSs specified for NG-RAN in section
and provide high-performance services to their subscribers II-B. In NG-RAN, the gNB comprises three units; the
[3]. In such an open environment, where multi-vendor gNB-Centralized Unit (gNB-CU), the gNB-Distributed Unit
software and hardware could co-exist on a shared platform, (gNB-DU), and the gNB-Remote Unit (gNB-RU). Each
MNOs will have the capability to configure their networks gNB-CU contains multiple gNB-DUs, and each gNB-DU
using advanced techniques, hence decreasing the Total Cost includes multiple gNB-RUs, where hardware or software-
of Ownership (TCO). In this network architecture, the based acceleration is a critical component for meeting the
O-RAN, known as the set of Next Generation Node B (gNB), extreme requirement for low latency operations that could be
comprises three components: operated in gNB-DU element. Fig. 1 demonstrates the RAN
1. O-RAN Central Unit (O-CU): is a centralized and virtual- architecture evolution and shows functional components,
ized element that is responsible for Packet Data Convergence terminologies, and interfaces among different components.
 3

open fronthaul link based on Split 6, which is particularly

Packet Core BBU
D-RAN
Backhaul well-suited to small cell, commercial networks. nFAPI covers
IP RRH
SCF’s established FAPI interfaces in a transport layer, enabling
EPC Backhaul
BBU Fronthaul open connections between components at PHY layer that can
C-RAN
IP CPRI/OBSAI RRH expand the architecture to the MAC layer, allowing it to be
Closed- virtualized and an open fronthaul link between a Small cell-
vEPC Backhaul vCU Midhaul DU Fronthaul
vRAN
IP F1 eCPRI RU RU (S-RU) and a Small cell-DU (S-DU). This enables MNOs
Open- Open-
to select these elements from several vendors if it results in
5GC Backhaul O-CU Midhaul O-DU Fronthaul
O-RAN the best solution for their particular use-case. The kind of
IP F1 eCPRI O-RU
network that will be used determines which splits will be
NG- NG-
5GC NG gNB-CU Midhaul gNB-DU Fronthaul chosen (among others). Split 6 does not need high-quality
NG-RAN
IP F1 eCPRI gNB-RU fiber for every link between RU and DU, whereas Split 7.2
requires it because PHY operations are concentrated in one
Fig. 1: RAN architecture evolution: functional components, location. This can be more practical and cost-effective in
terminologies, and interfaces. various enterprise contexts as it eliminates the need to purchase
and install optimized fiber to each RU and may even offer more
freedom to locate cells.
B. RAN Splits Standardization
RAN disaggregation and centralization, which relies on
The whole functionality of the gNB can be defined as adapting to particular FS, also enables virtualization of many
a chain of NFs [5], [6], where 3GPP [7] proposes eight NFs, with the software hosted on servers. In NG-RAN, the
different FS options for allocation of these functions among virtualized resources are shared dynamically among the gNBs
RAN components. The O-RAN offer different FS options and across gNBs’ components based on user needs. Thus, CU
for locating NFs. These FS options mainly rely on network is assumed to be fully virtualized, namely vCU. This allows
services and available transport links. Indeed each FS has its software development and implementation of network slicing
own set of requirements and, as a result, various alternatives in NG-RAN architecture. The next section will thoroughly
for connecting CU, DUs, and RUs have been recommended, discuss new technologies that can be adapted to the NG-RAN
each one with its own set of advantages and disadvantages. architecture.
Most commonly, option-2 is recognized as a FS between CU
and DU, whilst option-7 is perceived between DU and RU.
Various organizations are looking into the standardization of III. NG-RAN S OFTWARIZATION : ACTUAL T ECHNOLOGY
FSs to define the proper FS option. Nevertheless, there is no E NABLERS AND C HALLENGES
single, ideal FS agreed among standardization bodies as dif- Network Softwarization is a technology that allows a certain
ferent FSs will accommodate different applications. However, NF to run in software rather than hardware. The softwarization
it is unlikely that the industry will support all eight options. approach converts traditional network appliances with non-
Among the standardization bodies, 3GPP has supported Option standard hardware into software-based containers or Virtual
2 for highly centralized applications. Meanwhile, Option 6 Machines (VM) installed in standard equipment. In particu-
is being supported by SCF as the optimal FS for low-cost, lar, network softwarization is attributed to two main factors:
low-capacity deployments by focusing on its 5G network SDN and NFV. SDN facilitates rapid innovation through a
Functional Application Platform Interface (nFAPI) efforts (See programmable network and improves responsiveness, security,
Fig. 2. O-RAN Alliance recommends option 7.2 for networks efficiency, and cost-effectiveness. With NFV, the software runs
with high-capacity and high-reliability requirements. on low-cost hardware instead of running solely on costly
hardware, which translates into savings in on-site energy costs,
hence, reduction in the TCO.
gNB-CU 3GPP gNB-DU/RU
3GPP
Option 2
In this section, we will discuss the enabling technologies
(L3/High L2) (L2/L1)
F1 for NG-RAN softwarization such as virtualization and slicing,
S-CU 3GPP S-DU SCF S-RU
cloudification, and containerization.
SCF
(L3/High L2) Option 2 (L2/High L1) Option 6 (L1)
F1 5G-nFAPI

A. Network Slicing and Virtualization

O-CU 3GPP O-DU O-RAN O-RU
O-RAN
(L3/High L2) Option 2 (L3/High L2) Option 7-2 (Low L1) Network slicing and virtualization are promising technolo-
F1 eCPRI
gies in that the MNO of a network slice views the network
Fig. 2: The RAN split options recommended by 3GPP, SCF, slice as a distinct virtualized network shared with other net-
and O-RAN. work slices using NFV and SDN principles. MNOs fostered
virtualization initiatives and internal re-training, promoting
Fig. 3 shows eight 3gpp options along with their supported shifts to adopt this change within RAN architecture. Further-
standardization bodies. more, the Open Networking Foundation (ONF), a member of a
SCF’s nFAPI contributes significantly to the O-RAN plat- non-profit operator-driven consortium in the SDN movement,
form by offering the most deployable implementation of an has contributed its SDN platforms to O-RAN, expanding
 4

F1 5G-NFAPI eCPRI
3GPP SCF O-RAN

Option 2 Option 6 Option 7-2

Non-Real Time Functions (L3) Real Time Functions (L2) Real Time PHY Functions (L1)

High Low High Low High Low

IP RRC PDCP RF
RLC RLC MAC MAC PHY PHY

Network
Data Link Layer (L2) Physical Layer (L1)
Layer (L3)

Data

Option 1 Option 2 Option 3 Option 4 Option 5 Option 6 Option 7 Option 8

Central Unit (CU) Distributed Unit (DU) Remote Unit (RU)

Fig. 3: The 3GPP RAN splits: towards standardization

efforts developing RAN control standards called Software- different RAN slices, as different network slices require differ-
Defined RAN (SD-RAN) 1 . ent needs. For example, the requirements for handover support
Likewise, NG-RAN architecture can apply SDN/NFV to and mobility management for uRLLC slice are different from
decouple CU into a User Plane (UP) and Control Plane (CP) those for eMBB slice. The Management and Orchestration
part, where customized UP and CP processing can be tailored (MANO) can provide dynamic and adaptive slice management
to specific services such as scheduling multiple applications via an SDN controller by adopting a programmable SD-RAN
within PDCP layer, and control functions for a variety of [8] platform that separates CP and DP and allows a flexible
services that must be kept separate are hosted by CU-CP. CP design. However, as NG networks are expected to support
However, in DU, there is no CP and UP separation despite the m-RATs, RAN slicing solutions need to keep up with this
fact that MAC level scheduling has a significant impact on the concept. This adds another challenge, as it is questionable
RAN performance and service-specific aspects of scheduling if m-RATs could be multiplexed over the same potentially
and selection of PHY layer functionality and that the F1 specialized hardware or if each RAT requires its own dedicated
midhaul is unable to operate the DU on a real-time. The hardware.
virtualization could shift further to DU, namely as vDU, to
support the aforementioned aspects in the NG-RAN.
B. Cloudification and Containerization
Given the importance of a service-specific design of RAN
architecture, the traditional one-size-fits-all framework to mo- Cloudification and containerization are two other important
bile network infrastructure is incapable of dealing with the technologies that NG-RAN will embrace to define how NFs
presumed variety of services and widely differing NG re- and computing infrastructures should be distributed among
quirements [1]. To efficiently serve this traffic, virtualization RAN elements.
and slicing emerge as essential components within NG-RAN Cloudification adopted virtualization technology to effi-
architecture to create on-demand isolated slices for different ciently manage resources and services. The cost-effective
types of supported services such as eMBB, uRLLC, and administration of computing resources in RAN is made pos-
mMTC. However, implementing network slicing and virtual- sible by advancements in server virtualization. In particular,
ization causes additional complexities that need to be adopted cloudification enables the use of on-demand computing re-
into RAN architecture. sources in CU in the form of VMs instances or containers
Challenges: Creating and managing several RAN slices in rather than developing in physical infrastructure. Furthermore,
an isolated way is challenging as it has the disadvantage of containerization provides an Operating System (OS)-level of
inefficient usage of radio resources and needs the fast set- virtualization, where a single OS kernel can create multi-
up of new services or modification of existing ones. One ple isolated user spaces and run different applications from
solution is to provide efficient and dynamic spectrum-sharing CU. Accordingly, network functionalities are split down into
radio resources. However, this increases the need for shared or smaller micro-services and delivered as containers in CU using
coordinated scheduling mechanisms between slices and access the cloud-native method. In this regard, the most popular
channels while ensuring resource isolation (a customized FS containerization platform, Docker2 can create, deploy and
per slice) to fulfill service requirements is challenging. Another manage different micro-services to increase density, scaling,
issue is related to mobility management and response time of deployment speed, portability, and decrease cost.
1 [Link] 2 [Link]
 5

NG-RAN cloudification/containerization is indeed acting contribution to SCF Open RAN because it supports open
as a superior backend for applications running on accessing APIs between different components in a chipset to support
devices. In this way, it provides both manufacturers and end- interoperability between hardware components that implement
users the foundation for NG network slicing, on-demand provi- 3GPP PHY layer functions and the higher layer functions that
sioning of virtual resources, self-service functionality, resource run in software.
pooling, network hypervisors, scalability, high flexibility, a This standardization in NG-RAN qualifies vCU and vDU,
smooth evolution to support future standards, and easy support to distribute and scale resources in an isolated manner. For
for new revenue-generating services. For example, it enables instance, a non-virtualized DU may suffer critical load im-
placing MEC inside RAN architecture that provides ultra-low balance in some circumstances, e.g., distinct DUs deployed
latency, reliability, and scalability to accommodate massive to rural and urban areas or anchor versus non-anchor DUs
IoT devices. under a NR-DC scenario. Through vDU standardization, one
Challenges: Nevertheless, it creates a completely new in- can selectively allocate resources to each DU when required,
frastructure, which adds new complexities. For instance, fewer minimizing the possibility of load imbalance. This leads to
active hosts might aggregate more network links, which affects a reduction in resource usage due to statistical multiplexing
network latency. Deploying VNF redundancy can also have an gains, i.e., there is less need to supply emergency overload
impact on reducing network latency, which calls for a cost- capacity separately for each DU. The efficient resource sharing
effective approach. Another significant problem is effectively due to flexible migration (i.e., VNF scale in/out) [10] and
spreading the host’s computing power among the VNFs. Fur- dimensioning are other benefits of vDU standardization that
thermore, cloudification/containerization adds a new level of we will detail the potential benefits in this section.
complexity for MNOs, such as managing additional software
upgrades to ensure compatibility, particularly in multi-vendor A. On-Demand Scale In/Out
scenarios with various software roadmaps.
Container technology, in addition to CU/DU split and CU-
Hence, despite the challenges in vDU standardization, it
CP/CU-UP split can improve scalability, flexibility, and re-
stems important benefits towards an efficient RAN softwariza-
source efficiency. Each micro-service in the vCU and vDU
tion in NG-RAN, which we further discuss in the next section.
can have its own flavor (size) for flexible dimensioning. For
example, each micro-service of vDU can be scaled out if
IV. V DU S TANDARDIZATION : B ENEFITS AND additional cells are deployed. In vCU, traffic loads for CP
C HALLENGES and UP are balanced between each plane’s micro-service
separately to maximize resource usage efficiency. Each micro-
The vCU and vDU of NG-RAN are assumed by industry
service can be scaled on-demand or automatically based on
partners and academics to be implemented as fully virtualized
the current load status. In this regard, VNFs of vDU that are
VNFs, that might improve the performance of the NG-RAN
detached from the hardware resources can run on a device
architecture, enhance the implementation of service-aware
with COTS server. In vDU, the PHY and MAC layers demand
RAN slicing, deploy RAN slice subnets, simplify the network
extremely complicated computational tasks such as scheduling
operation, etc. The low layers are implemented on dedicated
algorithms and Forward Error Correction (FEC). These tasks
hardware, namely RU, which is generally realized as Physical
may cause a burden on a COTS server’s processing power and
Network Function (PNF). Virtualization is shifting from CU to
reduce vDU performance. As a result, some computationally
DU from dedicated hardware to software components due to
demanding processes with repeating structures, like FEC,
the limited capacity of each DU and the static barrier between
could be offloaded to different hardware chips for acceleration
each piece of hardware. For instance, it is anticipated that
and implemented as an option on a COTS server.
virtualizing DUs will require offloading of an NF to hardware
The MANO of vCU containers can be supported by the
(or software) accelerators but not CUs. Indeed virtualizing the
service orchestration tools such as Kubernetes (K8s) and its
CU and exposing it as a VNF is rather straightforward (consid-
lightweight versions (e.g., K3S, KubeEdge) for vDU, which
ering additional features like accelerating I/O data transfers).
are open-source container orchestration systems for automated
However, this is not the case for DU, where demanding High-
deployment of services, featuring high flexibility and scalabil-
PHY processing imposes additional challenges; thus, hardware
ity. Fig. 4 shows the virtualized architecture of vCU and vDU
or software acceleration for compute-intensive processing is
using containerization technology [11].
expected to fulfill RAN deployment requirements. Moreover,
vDU is recognized as a hardware-agnostic solution with its
components operating regardless of hardware requirements. B. PDCP as an enabler for DC and MEC Placement
This allows the fully virtualized RAN architecture employ- The presence of DC in NG-RAN is one further reason for
ing VNFs to benefit from flexible scalability and consistent vDU standardization. To promote DC without DU and CU
evolution. Accordingly, VNFs can run on lightweight servers split principles, devices could connect to two distinct gNBs
or edge cloud nodes that are implemented on COTS hardware, (CU and DU), but only one of the two DUs, the anchor DU,
which can interoperate with RUs from any vendor. Implement- is in charge of computing the split data streams via PDCP.
ing nFAPI support in NG-RAN software stack simplifies ven- Accordingly, the PDCP anchor DU receives a disproportionate
dors to base their system-on-chip designs on open platforms amount of the PDCP load, leading to improper resource con-
at every layer and supports Split 7 [9]. nFAPI is an important sumption between the PDCP anchor DU (over-utilized) and the
 6

gNB-CU
vCU-CP vCU-UP
that MEC-enabled vDU facilities this provisioning by hosting
Common
Common Micro-
CommonMicro- E1 Common
Common Micro-
Micro-
Common
V2X Application Server (AS) at the far-edge very close to
Services
Services Services
Services
Micro-services Micro-services where data is produced (by traffic cameras or sensors) so that
Kubernetes multi-cast data dissemination and mobility management pro-
(K8s)
cedures can be coordinated with lower latency. Nevertheless,
CP CP UP UP
Micro-Service 1 Micro-service n Micro-service 1 Micro-service n as this necessitates moving services from one edge server to
Container Engine another in response to a vehicle’s mobility, it adds another
Host OS
level of complexity to the MEC design. As mentioned above,
COTS Server
for deploying MEC on RAN side, we need to have a PDCP
functionally as an enabler and need to already decide where
(a) vCU Architecture to accommodate MEC; to edge side in vDUs close to vehicles
gNB-DU to meet hard delay and MEC computational requirements
vDU
Common
Common Micro-
Micro-
of vehicles or deploy in vCU to better coordinate vehicles
Common
Services
Services
Micro-services with tolerable delay and ultra bandwidth and higher MEC
Lightweight computational requirements. In order to jointly manage this
Kubernetes
(K3s) complexity and support the strict and wide-ranging require-
DU DU
Micro-Service 1 Micro-service n
ments of V2X services, MNOs must design efficient and
Container Engine
flexible network paradigms. This is particularly important for
Host OS
autonomous driving and traffic safety applications to prevent
accidents and optimize traffic flow.
COTS Server

(b) vDU Architecture C. Joint Slicing vs Multiple splits

Fig. 4: The containerized vCU and vDU Architecture [11]. Up to now, only using a single FS is allowed to be
implemented in gNB components. There is a need to support
and standardize multiple FSs per unit to meet NG service
non-anchor DU (under-utilized). As stated in 3GPP Release 15 requirements. The NG RAN slicing is a complementary tech-
and 16 [4], [7], [12], systems can exploit PDCP duplication nology that enables running different FSs per unit (i.e., vCU
to improve data reliability. Hence, two replicas of an identical or vDU), thus guaranteeing functional isolation between slices
PDCP Protocol Data Unit (PDU) may be sent. The secondary that is crucial as it enables a slice-custom FS within a single
RLC entity could be assigned to a different DU referred to unit [15]. For instance, in a given gNB, for a tele-driving
as DC-based PDCP duplication, or to the same DU known service that requires a strict latency (10 ms), a slice using
as Carrier Aggregation (CA)-based PDCP duplication. With virtual PDCP in vDU (close to vehicle) with reduced HARQ
multiple independent transmission paths, DC in vDU mitigates delay can support this service. On the other hand, for traffic
load imbalance, improves reliability, and decreases latency efficiency service, which has tolerable delay requirements, a
which is more desirable for uRLLC services. Furthermore, the slice better suits with a centralized FS with PDCP in vCU such
data splitting in diverse simultaneous traffic flows boosts the that neighboring cells can coordinate more accurately and the
network capacity for eMBB services. NG-RAN is anticipated handover process can be made simpler for these services. (See
to meet a variety of demands coming from massive devices, Fig. 5).
low latency services (such as uRLLC), and vertical industry
demand for higher capacity in order to achieve lower latency V. C ONCLUSIONS
and improved performance. Deploying NFV and SDN in the This study presents the gains of NG-RAN softwarization
RAN, which will virtualize all functions and resources of along with vDU standardization benefits for MNOs to adopt
the RAN elements (vDU, vCU), is a promising solution to cutting-edge technologies and operate their network more
meet these demands. Another potential solution is deploying flexibly and efficiently. We discuss both challenges and gains
MEC on RAN side next to users. Referring to [13] the of NG-RAN softwarization that leads to convenient hardware
placement of PDCP function is conceived as the bottleneck maintenance and maximized lifecycles. To optimize resources
for accommodating a MEC service, implying the baseline FS such as CPU core and memory, vDU allows MNOs to effi-
option required to support MEC either in vCU or in vDU. ciently allocate resources as well as automatically scale vDU
Hence, the MEC accommodation relies on the location of components in response to changes in network traffic patterns.
PDCP function. In that sense, resource pooling can further improve resource
The major effect of vDU standardization for uRLLC ser- efficiency since virtualization eliminates hardware boundary
vices will be at the network edge, where low-latency access to restrictions.
cloud-like resources is possible. Vehicle-to-everything (V2X) MNOs can upgrade their networks without outages using
application is a primary use-case of uRLLC that has been software that is produced and deployed on time. Without mod-
identified by 3GPP [14], which will enable communication ifying hardware, novel technologies could indeed be initiated
among vehicles and infrastructure. V2X service provisioning and network capacity could be expanded. Yet, it adds com-
necessitates a wide range of data rates, reliability, and latency plexity to managing software upgrades that may need frequent
 7

Edge Cloud Far Edge NR NG

VNFs VNFs PNFs Services

VR and 3D video
MAC UPF MEC
eMBB Slice
PHY RF
RLC PDCP

V2X Services
RRC RLC UPF MEC PHY
V2X/uRLLC
Slice V2X
IP MAC PDCP RF
AS

vCU vDU RU UE

Fig. 5: The Schematic Overview of vDU Standardization Benefits

and periodic upgrades for any repair and maintenance. We [11] Samsung, “Virtualized Radio Access Network: Architecture, Key tech-
demonstrate that network planners can manage their systems nologies and Benefits,” Tech. Rep., 2019.
[12] 3GPP, “NR; Packet Data Convergence Protocol (PDCP) Specification,
more successfully than in earlier hardware-based solutions document 3GPP, TSG-RAN 38.323 V15.5.0 ,” 2019.
thanks to the functional independence of the vDU and vCU. [13] ETSI, “ETSI “Multi-Access Edge Computing”, Online:
Network efficiency is increased by the vCU’s ability to be [Link]
edge-computing, ,” 2017.
installed on-site with a vDU, in a central location, or near the [14] 3GPP, “Study on enhancement of 3GPP support for 5G V2X services,”
network edge, connected to the vDUs of several sites. Tech. Rep. TR 22.886 V1.0.0,” 2016.
[15] B. Ojaghi, F. Adelantado, A. Antonopoulos, and C. Verikoukis, “Impact
of network densification on joint slicing and functional splitting in 5g,”
ACKNOWLEDGMENT IEEE Communications Magazine, vol. 60, no. 7, pp. 30–35, 2022.
This work has been partially supported by the SPOTS
project (RTI 2018-095438-A-I00), the FEM-IOT (001- B IOGRAPHIES
P-001662), the RF-VOLUTION (PID2021-122247OB-I00),
BEHNAM OJAGHI (bojaghi@[Link]) is Marie
MARSAL (101017171), OPTIMIST (872866), and 5G-ERA
Sklodowska-Curie Alumni from Universitat Oberta de
(101016681).
Catalunya (UOC). He is doing a Ph.D. at Polytechnic
R EFERENCES University of Catalunya (UPC).
FERRAN ADELANTADO (ferranadelantado@[Link]) is
[1] I. Afolabi, T. Taleb, K. Samdanis, A. Ksentini, and H. Flinck, “Network
Slicing and Softwarization: A Survey on Principles, Enabling Technolo- Associate Professor at UOC and an Adjunct Associate Pro-
gies and Solutions,” IEEE Communications Surveys Tutorials, p. 1, 2018. fessor at Universitat de Barcelona (UB).
[2] C. Mobile, “C-RAN: The Road Green RAN White Paper V3. 0,” 2011. CHRISTOS VERIKOUKIS (cveri@[Link]) re-
[3] O-RAN, “O-RAN Working Group 1, “O-RAN Architecture Description
- v2.00,” Technical Specification,” 2020. ceived his Ph.D. degree from UPC in 2000. He is Associate
[4] 3GPP, “NR; Overall description; Stage-2. Technical Specification TS- Professor at Patras University.
38.300, v16.1.0,” 2020.
[5] Small Cell Forum, “Small cell virtualization functional splits and use
cases,” 2016.
[6] L. M. P. Larsen, A. Checko, and H. L. Christiansen, “A survey of
the functional splits proposed for 5g mobile crosshaul networks,” IEEE
Communications Surveys Tutorials, vol. 21, no. 1, pp. 146–172, 2019.
[7] 3GPP, “TR 38.801. Study on new radio access technology: Radio access
architecture and interfaces (Release 14),” 2017.
[8] X. Foukas, M. K. Marina, and K. Kontovasilis, “Orion: RAN Slicing
for a Flexible and Cost-Effective Multi-Service Mobile Network Archi-
tecture,” in Proceedings of the 23rd Annual International Conference on
Mobile Computing and Networking, ser. MobiCom ’17. ACM, 2017.
[9] S. C. Forum, “Small Cell Open RAN: A catalyst for new 5G business
models, An SCF guide to small cell Open RAN architecture and
deployment, SCF239.10.01,” 2022.
[10] S. Ramanathan, A. Bhattacharyya, K. Kondepu, M. Razo, M. Tacca,
L. Valcarenghi, and A. Fumagalli, “Demonstration of containerized
central unit live migration in 5g radio access network,” in 2022 IEEE
8th International Conference on Network Softwarization (NetSoft), 2022,
pp. 225–227.

View publication stats