See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/357621849

Edge Computing and the Fifth Generation (5G) Mobile Communication: A

Virtualized, Distributed System Towards a New Networking Concept

Conference Paper · December 2021

CITATIONS READS

12 1,263

3 authors:

Rekha Reddy Shaden Baradie

Deutsches Forschungszentrum für Künstliche Intelligenz 4 PUBLICATIONS 50 CITATIONS
26 PUBLICATIONS 73 CITATIONS
SEE PROFILE
SEE PROFILE

Christoph Lipps
Deutsches Forschungszentrum für Künstliche Intelligenz
97 PUBLICATIONS 427 CITATIONS

SEE PROFILE

All content following this page was uploaded by Christoph Lipps on 06 January 2022.

The user has requested enhancement of the downloaded file.

 Edge Computing and the Fifth Generation (5G)
Mobile Communication: A Virtualized, Distributed
System Towards a New Networking Concept
Rekha Reddy† , Shaden Baradie† and Christoph Lipps†∗
† Intelligent Networks Research Group, German Research Center for Artificial Intelligence
D-67663 Kaiserslautern, Email: {firstname.lastname}@dfki.de
∗ Institute for Wireless Communication and Navigation, University of Kaiserslautern

D-67663 Kaiserslautern, mail: {lastname}@eit.uni-kl.de

Abstract—With the fast-growing wireless communication tech- which was a motivation to find a better and more efficient
nologies, Beyond 5G (B5G) and the Sixth Generation (6G) solution with reduced costs.
wireless systems empower the Internet of Things (IoT) and The concept behind virtualization is incorporating the ele-
edge computing. Hence, the transition towards the virtualized
edge distributed framework enables applications to operate with ments of flexibility and intelligence into the network infras-
higher performance and efficiency by merging people, data and tructure by transferring it into a virtual environment that can be
things. In this work, the distributed system of virtualized 5G hosted by less physical hardware. One of the advantages of the
Wireless communication and Mobile Edge Computing (MEC) is virtualized network infrastructure is its ability to hold multi-
constructed over a shared resource. This system is subjected tenancy and multivendor equipment. In addition, it decreases
to varying load through real-time application scenarios with
a simulated User Equipment (UE) and operational Nextcloud the costs of installation and maintenance and opens up more
instances over MEC and their performance metrics are analyzed possibilities for the scientific communities to hold various
by scaling infrastructure. During the huge load of network traffic, experiments and testing in a more simplified environment.
the system is auto-scaled to provide high availability. Considering the previous qualities and more, virtualization
Index Terms—Radio Access Network (RAN), Edge Computing, has already become part of the new networking concept of
User Equipment, Virtual Network Functions (VNF)
5G, and in the future, B5G, and 6G, by virtualizing most
of the network functions of the core network and the Next-
I. V IRTUALIZATION TOWARDS A NEW NETWORKING
Generation Radio Access Network (NG-RAN) architecture,
CONCEPT
and introducing Network Function Virtualization (NFV) [1].
The first breakthrough in mobile and wireless communica- At the moment, NFV is still under research and examination
tions –which was introducing voice communication through to be optimized for better management and orchestration
the first generation analogue cellular system– had paved the between the different Virtual Network Functions (VNFs) in
way for continuously developing new mobile networks gener- the shared infrastructure. Therefore, integrating NFV with
ations all the way through 2G, 3G, 4G (LTE), and the Fifth Artificial Intelligence (AI) tools and Machine Learning (ML)
Generation (5G), until recently reaching Beyond 5G (B5G) algorithms is considered as a promising solution. By exploiting
and the beginnings of the Sixth Generation (6G). 6G proposes NFV into various aspects of the network management, such
and promises new cutting the edge technologies that will as Software-Defined Networking (SDN) and network slicing,
change the human life as we know it, such as, among others, they will be able to achieve high flexibility and better service
Holographic-Type Communication (HTC), Extended Reality modularity effectiveness [1]. Furthermore, the concept can
(ER), Tactile Internet, Multi-Sense Experience, Digital Twin be extended to combine SDN and Physical Layer Security
[1]. (PhySec) to satisfy the network security requirements [2].
In order to meet the ever-increasing demand and ambitious This work aims to highlight the advantages and capabilities
future vision, the developments of a new generation usually of virtualization by implementing two different technologies,
start simultaneously while the previous one is still under such as 5G and MEC on different virtual machines hosted
improvement. One of the main challenges in the development by the same physical hardware resources. The performance
process of a new generation is its compatibility with the of the hosted machine is analyzed during the application’s
existing infrastructure and to what extent the infrastructure run-time to get an evaluation of the virtualization’s efficiency.
has to be altered to serve the new generation and its services. Edge computing paradigm is currently a hot topic in the
The requirement for infrastructure modification, addition or field of research, and it is considered as a key enabler for
maintenance always comes with huge costs and sometimes virtualization services. Therefore, the applications of choice:
with unexpected impact on the already existing elements, Free5GC mobile core networks and Nextcloud edge server
 environment, 5G relied on the following new technologies to
be able to meet the previous demands: mm wave communica-
tion, small cells, and massive Multiple Input Multiple Output
(mMIMO) [3].

A. Edge Computing: Characteristics & Models

In order to serve the increasing QoS, low-latency communi-
cation, and wide bandwidth demands of the highly-interactive
applications of 5G, a new distributed computing paradigm was
adopted. Edge computing was proposed as an extension to
the centralized computing paradigm; cloud computing. Edge
computing brings the cloud closer to the end-used, which leads
to improving the cloud’s capabilities and overcoming some of
its main limitations, such as latency, security and the limited
processing and computation capabilities [3].
In addition to the previous advantages of edge computing
over cloud computing, such as low-latency, the following
characteristics make edge computing stand out even more [4]:
Fig. 1: The implementation design of the 5G system and edge • Dense Geographical Distribution: the dense distribution
server. of the edge network’s infrastructure enables the network
administrator to provide location-based mobility services
without the need to transverse the whole network. Be-
for file hosting services, were implemented to represent the sides, the ability to perform fast and more accurate big-
concept of virtualization on edge as shown in Figure 1. data and real-time analytics.
The rest of this work is structured as follows: Section • Mobility support & location awareness: edge computing
II provides a brief overview of edge computing and the supports direct communication with mobile devices by
architecture of 5G with a glimpse of its elements including decoupling the host identity from the location identity
Radio Access Networks (RAN), and how RAN can be brought using Locator ID Separation Protocol (LISP). It also
closer to the end-user by incorporating edge computing and enables the end-users to benefit from the services of the
virtualization. Section III illustrates the implementation design edge closest to their location. This feature is valid through
structure and the properties of its components, including deploying a location identification technology into the
both the physical and virtual parts of the infrastructure. The end-device, such as GPS.
performance analysis of the virtual machines is represented • Context-Awareness: edge computing can use the context-
in Section IV through observing the Central Processing Unit awareness feature of mobile devices to make better of-
(CPU) usage and network traffic variations during real-time floading decisions. It also enables the service provider to
applications, followed by a discussion for further improve- use this information to provide the user with a higher
ments and possible extensions. Finally, Section V gives a brief Quality of Experience (QoE).
overview and summarizes the whole work with proposals and • Heterogeneity: the heterogeneity of the edge computing
suggestions for future work. paradigm in different aspects from end-devices to edge
servers and networks, resembles the heterogeneous nature
II. OVERVIEW OF 5G W IRELESS C OMMUNICATION AND of the 5G environment. Which assures its compatibility
E DGE C OMPUTING with 5G and its ability to meet the applications’ require-
ments.
The Fifth Generation introduced new cutting-edge applica-
tions, such as, Augmented Reality (AR), Virtual Reality (VR), Cloudlets, Fog computing, and Mobile Edge Computing
tactile internet, etc. Therefore, the new generation was based (MEC), are three different models of edge computing, each
on three main objectives, namely classified into: enhanced one of them is located in a different location in the network
Mobile Broad-Band (eMBB) which provides reliable broad- and provides different services depending on its characteristics
band access over wide coverage areas, Ultra-Reliable Low- and features.
Latency Communication (URLLC), which enables end-to-end • Cloudlets: similarly to cloud computing, cloudlets aim
communications with less than 5 ms latency, and massive to enable computing-intensive tasks by bringing parts of
Machine Type Communication (MMTC), that supports the the cloud, mostly computing resources and data storage,
wireless connectivity of a huge number of devices. closer to the end-user. Cloudlets are designed to be self-
Due to the ever-increasing number of mobile users each managing and resource-rich systems, which gives them
year, the high Quality of Service (QoS) requirements, and the an advantage over the traditional cloud in terms of high-
heterogeneous nature of the future wireless communication’s speed access and low latency [5][6].
 • Fog Computing: due to its heterogeneous nature, fog • Network Repository Function (NRF) adds a layer of load-
computing enables the execution of real-time applications balancing by storing a profile of each NF and allowing the
at the edge of the network by using small access points new NFs to discover which services they can be provided
known as Fog Computing Nodes (FCNs). The fog layer, with by accessing instances from a different NF.
including its various types of FCNs, such as switches and • Policy Control Function (PCF) provides a policy frame-
routers, can be located freely anywhere between the cloud work to influence the network’s behavior and decides on
and the end-users [6]. the suitable QoS parameters for the various applications
• Mobile Edge Computing (MEC): the distributed nature and services provided by each network slice.
of Mobile Edge Computing reduces the computation and • Unified Data Management (UDM) performs different
storage offloading between the end-users and the core authentication and subscription management mechanisms
network, by enabling the storage and processing services based on the specific requirements of each component in
at the end-users’ level. Which in turn results with lower the network slice.
latency and higher quality of service. Therefore, real-time • Network Exposure Function (NEF) permits the exposure
applications, such as content sharing, traffic management, of capabilities between different NFs.
and E-health, can be enabled through the deployment of • Session Management Function (SMF) is responsible for
MEC services [7]. initiating and managing user sessions, interacting with
the User Plane Function (UPF), and supporting different
B. 5G System Architecture protocol models.
As illustrated in Figure 2, the 5G system architecture • Access and Mobility Management Function (AMF) acts
consists of three main elements: Core Network (CN), Ac- as an access point for the UE and as a connection bridge
cess Network (AN), and UE. The core network is divided between RAN and CN. It then determines the suitable
into Control Plane (CP) and User Plane (UP) with different SMF for the user session management depending on the
Network Functions (NFs). Unlike the previous generations, required services.
the 5G system has a Service-Based Architecture (SBA) as • Authentication Server Function (AUSF) is responsible for
it incorporates network slicing and virtualization techniques enabling AMF to authenticate the UE and access the CN’s
through the network’s domains to be able to provide various services.
services over the network and meet the users requirements. • User Plane Function (UPF) enables the connection be-
tween the UE and external networks.
• Application Function (AF) can be used for influencing
and customizing the network slice capabilities.
• Unified Data Repository (UDR) provides data storage
services for other NFs.
The evolution of RAN: RAN had continuously evolved
through different mobile networks generations in several as-
pects: from its layout and architecture to its functionality and
main role in the wireless communication system. Starting
from the first basic RAN that consists only of a Radio
Unit (RU), responsible for transmission and reception, and
a Processing Unit (PU), responsible for radio management,
resource sharing, and encryption. The number of users was
much less at the time with limited services available, therefore,
Fig. 2: 5G system architecture consisting of three main ele- the basic RAN was sufficient for its purpose. However, as the
ments: core network, access network, and external network. wireless communication evolved, with more services to offer
and a wider spectrum range, the number of users increased
5G core network: As part of the SBA, the CN is able to significantly, and new technologies and frameworks were
provide different services including authentication, security, introduced, such as densification, frequency reuse, and later
session management, and traffic aggregation using virtualized on millimeter wave (mm-wave) in 5G networks [10].
NFs. The inter-connection between the VNFs enables exploita- The traditional infrastructure of RAN had to be evolved as
tion and sharing of data throughout the network. Each NF well to meet the new services’ requirements. Some main de-
holds a specific responsibility in the communication process velopments in RAN overtime that were proven most effective
as follows [8][9]: are highlighted briefly as follows [10]:
• Network Slice Selection Function (NSSF) is responsible • From Base Station (BS) centric to UE centric: instead
for the slice decision-making process when a new UE of assigning the BSs based on the signal’s strength, a
joins the network. It makes the decision depending on UE can receive a signal from multiple BSs based on
which slice will be able to serve the new UE optimally. a coordinated work between them, which reduces the
 interference between signals from different BSs. TABLE I: Hardware Specifications.
• mmWave and Beamforming: communication in mmWave Specifications
spectrum was introduced in 5G systems as a solution Host Hardware Intel(R) Core(TM) i9-10885H
for the high data rate requirements and the limited RAM 32GB
bandwidth. It was integrated with beamforming from CPU Cores 8
Logical Processor 16
the antennas directly into the targeted UE to avoid the Frequency 2.40GHz
attenuation and diffraction that can be an issue in higher Operating System Windows 10
Virtual Operating System Ubuntu 18.04
frequency bands.
• From single-point to multipoint transmission: Cloud Ra-
dio Access Network (CRAN) has adopted the UE centric one is a 5G system, where the open-source Free5GC project
approach to avoid inter-cell interference at the edge user, [12] was used for creating two VMs: one operates as a mobile
where each UE is being served by a cluster of BSs. core network and user Subscriber Identity Module (SIM)
• Coordinated Transmission: a type of transmission, known
registration, while the other has RAN and a simulated UE.
as, Coordinated Multipoint (CoMP) where multiple trans- Free5GC was deployed aiming to examine the operability,
mitting points can serve each user. To avoid intra-cluster functionality, and connection of the virtual elements of 5G
interference between the BSs, the transmitting points are network. The second application is creating an edge server
coordinated using different techniques, such as coordi- using a Kubernetes cluster environment. This edge server is
nated beamforming, distributed transmission, and joint able to store data on the edge closer to the end-user, by using
transmission. a client-server software called Nextcloud. Following a brief
Virtual Radio Access Network: Since RAN provides a radio overview of the functionality and features of each application:
connection bridge between the UE and the CN, it can be con-
• Free5GC is an open-source project for 5G mobile net-
sidered as one of the regions where latency originates, which
works. This project aims to implement the standalone 5G
makes integrating edge computing low-latency communication
core network (5GC) features based on the 3GPP Release
with RAN a possible solution for more reliable and faster
15 (R15) and beyond. Each one of the core network’s
communication. One way of deploying RAN at the network’s
functions is packed individually as a VNF [12]. Similarly
edge is virtualization, or the so-called Virtual Radio Access
to Figure 2, the network’s architecture in Free5GC is
Network (vRAN). The virtualization of RAN aims to add a
divided into CP and UP, each with its own functions,
layer of flexibility to it, enabling seamless adjustments in case
such as SMF, UPF, etc. This separation enables each
of a change in the service requirements. Moreover, it also
plane to be deployed and developed independently of the
helps in reducing the total cost of the network considering
other and provides more flexible deployments of the NFs
that the deployment, management, and optimization of RAN
through NFV. In addition to incorporating the previously
is responsible for approximately 70% of it [11].
mentioned NFs in Section II, Free5GC uses MongoDB
III. 5G AND E DGE I MPLEMENTATION as a database for data storage [13].
• Nextcloud, a platform that offers file hosting services
In this proposed work, a 5G system was constructed to
on the cloud, was applied on a virtual machine using
operate alongside an edge server sharing common hardware
the container-management tool Minikube [14]. Similarly
resources as shown in Figure 1. The 5G system was subjected
to Kubernetes, Minikube provides management and or-
to varying load usage from a simulated UE. While the edge
chestration to a cluster of containers, however, Minikube
server was deployed with Kubernetes cluster to provide a high-
was a more suitable choice in our case since it helps in
performance Nextcloud application that can operate in real-
implementing local Kubernetes cluster, while at the same
time with a possibility for a multi-user environment.
time providing the Kubernetes features [15]. External
A. Physical Infrastructure clients from outside the cluster are enabled to connect
to the Nextcloud server that resides inside the cluster
The experimental setup consists of a physical hardware with
by implementing Kubernetes ingress object, which is
the host operating system and a virtualized infrastructure with
an Application Programming Interface (API) object that
the specifications as listed in Table I. The host machine was
manages external access to the cluster’s services [15].
able to contain a virtualized infrastructure using a hosting
Figure 3 shows the deployment and scaling model of
hypervisor. Oracle VirtualBox was installed over the host
Nextcloud instances. When the allocated CPU resources
machine and it was used to deploy a distributed system over a
are consumed above 80%, new instances are spawned and
virtualized layer as several separated Virtual Machines (VMs).
load balanced automatically.
The structural design of the implemented distributed system
is represented in Figure 1.
IV. D ISCUSSION AND U SE -C ASE A PPLICATION
B. Virtual Infrastructure The virtualized distributed system that was implemented in
As indicated, two applications were implemented on the this work operates with 5G system and an Edge computing
same hosting environment using Oracle VirtualBox. The first infrastructure.
 Fig. 5: 5G Core Network Metrics during the initial UE network
establishment connection and ping activity with average data
Fig. 3: Nextcloud Instances deployment and scaling through rate of 1Mb/s.
Kubernetes.

A. 5G Virtualized System
The 5G user equipment is subscribed through the available
web console in the 5G core component, and the different
instances of 5G core, RAN, and simulated UE were initialized.
After the connection, a new network interface (IP address)
was allocated to the UE in order to be controlled through this
interface. As a first use case, to test the internet connectivity
Fig. 6: 5G Core CPU utilization in Lubuntu. At a time instance
of the UE, the reachability to the network is validated through
of 07:58, the UE was initialized with the network interface and
ping command line utility. During this operation, the hardware
load consumption showed some variation during the streaming
resources such as CPU and network interface of the 5G core
of a high-quality video over the network.
system is monitored, and the resulting performance metrics
are shown in the Figures 4 and 5. From the metrics, the
consumption of the core system hardware resources can be
observed during the time of connectivity and while pinging to
the internet. The average data rate in this case is 1Mb/s.

Fig. 7: 5G Core Network Metrics during the initial UE network

establishment connection and the streaming of a high-quality
video over the network.

Fig. 4: 5G Core CPU utilization. During the initial 5 minutes,

the peaks are caused due to the system initialization, which user traffic and the hardware resources consumed. During the
is later followed with 20% of consumption during the UE high load usage, the instance can be scaled automatically and
network establishment connection and varying load for ping a new Nextcloud service is spawned. As a test case scenario,
activity. the instances of Nextcloud are manually scaled to operate
with an additional 3 replicas and the corresponding host CPU
As a second use case, the simulated UE was initialized over utilization is shown in the Figure 8. The overall time duration
Lubuntu, a minimalistic operating system, to test the effect of to create and initialize and to balance the load for 3 replicas
video streaming over the network interface that was allocated is 1 minute. The average network usage of Minikube interface
through the RAN. The test results are shown in the Figures 6 is shown in the Figure 9. The resulting network utilization
and 7. From the resource utilization metrics, it can be noticed increased to balance the operational load of the new replicas.
that the overall CPU consumption is lower than 15% and the The incoming connections are enabled and monitored through
maximum network data rate is 550 MB/s during the high- the ingress object. The deployed application is designed to
quality video streaming. operate on a single node cluster, however, it can be extended
and managed to operate heterogeneous applications through
B. Nextcloud a multi node environment. Kubernetes implemented architec-
Nextcloud is deployed as a service and made available ture benefits from high availability and scalability, therefore
through TCP/IP port 80. The load is balanced based on the enhancing the system performance. During an abrupt failure
of an instance, Kubelet automatically detects the failure and [2] C. Lipps, D. Krummacker, and H. D. Schotten, “Se-
respawns the new service, thus enabling high availability. curing Industrial Wireless Networks: Enhancing SDN
with PhySec,” in 2019 Conference on Next Generation
Computing Applications (NextComp), 2019, pp. 1–7.
DOI : 10.1109/NEXTCOMP.2019.8883600.
[3] N. Hassan, K.-L. A. Yau, and C. Wu, “Edge Computing
in 5G: A Review,” IEEE Access, vol. 7, pp. 127 276–
127 289, 2019. DOI: 10.1109/ACCESS.2019.2938534.
[4] W. Z. Khan, E. Ahmed, S. Hakak, I. Yaqoob, and
A. Ahmed, “Edge computing: A survey,” Future Gen-
eration Computer Systems, vol. 97, pp. 219–235, 2019.
Fig. 8: Nextcloud CPU utilization, manual scaling of instances DOI : 10.1016/j.future.2019.02.050.
with 3 additional replicas, resulting CPU usage is shown [5] U. Shaukat, E. Ahmed, Z. Anwar, and F. Xia, “Cloudlet
during the instance between 16:45 to 16:46. deployment in local wireless networks: Motivation, ar-
chitectures, applications, and open challenges,” Journal
of Network and Computer Applications, vol. 62, pp. 18–
40, 2016. DOI: 10.1016/j.jnca.2015.11.009.
[6] E. Krishnasamy, S. Varrette, and M. Mucciardi, “Edge
Computing: An Overview of Framework and Applica-
tions,” 2020.
[7] Y. Jararweh, A. Doulat, O. AlQudah, E. Ahmed, M.
Al-Ayyoub, and E. Benkhelifa, “The future of mobile
cloud computing: Integrating cloudlets and Mobile Edge
Fig. 9: Nextcloud Network Metrics, during the manual scaling Computing,” in 2016 23rd International Conference on
of the instances at the time 16:45. Telecommunications (ICT), 2016, pp. 1–5. DOI: 10 .
1109/ICT.2016.7500486.
[8] M. Shariat, Ö. Bulakci, A. De Domenico, C. Man-
V. C ONCLUSION AND O UTLOOK
nweiler, M. Gramaglia, Q. Wei, A. Gopalasingham,
The growth of the virtualized distributed systems had paved E. Pateromichelakis, F. Moggio, D. Tsolkas, et al.,
the way for merging heterogeneous systems by utilizing shared “Aflexible network architecture for 5G systems,” Wire-
hardware resources, and thus optimizing the deployment and less Communications and Mobile Computing, vol. 2019,
operational cost. This work provides a distributed system 2019. DOI: 10.1155/2019/5264012.
constructed over a virtualized infrastructure to support the [9] A. Banchs, D. M. Gutierrez-Estevez, M. Fuentes, M.
execution of 5G system and MEC to work simultaneously Boldi, and S. Provvedi, “A5G Mobile Network Archi-
using shared resources. The discussed MEC application pro- tecture to Support Vertical Industries,” IEEE Commu-
vides a system with high availability and scalability through nications Magazine, vol. 57, no. 12, pp. 38–44, 2019.
the application of Kubernetes. DOI : 10.1109/MCOM.001.1900258.
This work uses the concept of application deployment [10] S. K. Singh, R. Singh, and B. Kumbhani, “The Evo-
through virtual machines. This virtualization concept can be lution of Radio Access Network Towards Open-RAN:
further extended to operate in a containerized environment Challenges and Opportunities,” in 2020 IEEE Wireless
with light weighted instances of the applications, where in- Communications and Networking Conference Work-
dividual components can be created with flexible scaling, shops (WCNCW), 2020, pp. 1–6. DOI: 10 . 1109 /
management, and orchestration of the core components with WCNCW48565.2020.9124820.
varying network load. [11] S. Niknam, A. Roy, H. S. Dhillon, S. Singh, R. Banerji,
ACKNOWLEDGMENT J. H. Reed, N. Saxena, and S. Yoon, “Intelligent O-
This work has been supported by the Federal Ministry of RAN for beyond 5G and 6G wireless networks,” arXiv
Education and Research of the Federal Republic of Germany preprint arXiv:2005.08374, 2020.
(Förderkennzeichen 16KIS1283, AI-NET PROTECT). The [12] Free5GC, https://www.free5gc.org/, Jan. 2019.
authors alone are responsible for the content of the paper. [13] C.-W. Liao, F. J. Lin, and Y. Sato, “Evaluating NFV-
enabled Network Slicing for 5G Core,” in 2020 21st
R EFERENCES Asia-Pacific Network Operations and Management Sym-
[1] W. Jiang, B. Han, M. A. Habibi, and H. D. Schotten, posium (APNOMS), 2020, pp. 401–404. DOI: 10.23919/
“The Road Towards 6G: A Comprehensive Survey,” APNOMS50412.2020.9236978.
IEEE Open Journal of the Communications Society, [14] Minikube, https://minikube.sigs.k8s.io/docs/.
vol. 2, pp. 334–366, 2021. DOI: 10 . 1109 / OJCOMS . [15] Kubernetes, https : / / kubernetes . io / docs / concepts /
2021.3057679. overview/.

View publication stats