IBNSlicing: Intent-Based Network Slicing Framework for 5G Networks using Deep Learning

Abstract

Network slicing is an important pillar of 5G networks that empowers the network operators to provide the different quality of services (QoS) to the users. It enables network operators to split the physical network into multiple logical networks to meet different QoS requirements. In this research paper, we have designed an intent-based network slicing framework that can slice and manage the core network and radio access network (RAN) resources efficiently. It is an automated system, where users just needs to provide higher-level information in the form of intents/contracts for a network slice, and in return our system deploys and configures the requested resources. Moreover, a deep learning model Generative Adversarial Neural Network (GAN) has been used for the management of network resources. Several tests have been performed by creating three slices with our system, which shows better performance in terms of bandwidth and latency.

Full text

IBNSlicing: Intent-Based Network Slicing
Framework for 5G Networks using Deep Learning
Khizar Abbas, Muhammad Afaq, Talha Ahmed Khan, Asif Mehmood, Wang-Cheol Song*
Department of Computer Engineering, Jeju National University,Jeju-si, South Korea
khizar.abbas@jejunu.ac.kr, philo@jejunu.ac.kr, afaq@jejunu.ac.kr
Abstract—Network slicing is an important pillar of 5G net-
works that empowers the network operators to provide the
different quality of services (QoS) to the users. It enables
network operators to split the physical network into multiple
logical networks to meet different QoS requirements. In this
research paper, we have designed an intent-based network slicing
framework that can slice and manage the core network and radio
access network (RAN) resources efficiently. It is an automated
system, where users just needs to provide higher-level information
in the form of intents/contracts for a network slice, and in
return our system deploys and configures the requested resources.
Moreover, a deep learning model Generative Adversarial Neural
Network (GAN) has been used for the management of network
resources. Several tests have been performed by creating three
slices with our system, which shows better performance in terms
of bandwidth and latency.
Index Terms—E2E Network Slicing, 5G Networks, OSM,
FlexRAN, Deep Learning
I. INTRODUCTION
It is expected that the fifth-generation mobile network (5G)
can cover a wide range of industrial use cases and improve net-
work performance significantly. The current mobile networks
are no more efficient to handle the multi-service demand and
provide the different quality of services (QoS) to their users
[1]. The virtualized and programmable nature of the 5G mobile
network is the best and cost-effective solution for ensuring on-
demand services. Network slicing is an important feature of the
5G network that allows you to create multiple isolated logical
networks over the same physical infrastructure, is referred to
as a slice, for the mission of supporting multiple use cases,
with diverse service requirements. The network slicing in a
5G network is achieved using network function virtualization
(NFV) and software-defined network (SDN) technologies [2],
[3].
5G network architecture is designed on the service-oriented
pattern that can provide better and different QoS to users for
different kind of applications scenarios like data transmission
and low latency, etc. The 5G network should be able to provide
all kind of services to every kind of users requirements such
as in the highly-dense area where the cellular traffic is very
high, in those cases which required seamless service with high
bandwidth, indoor environment, 5G networks should be cover
these cases and provide seamless high bandwidth even with
fast speed mobility [4], [5]. On the other hand, in the case of
a large amount of widespread low power-constrained devices
(sensors, smartphones, security cameras, etc.) that needed a
reliable connection, 5G networks assure the connectivity of
millions of these low power and constrained resources device
at very low cost. The other use-case is connected vehicle, smart
factories, real-time applications (health care monitoring) which
required ultra-low latency, 5G networks fulfill the requirements
of low latency application and provide secure communications
by introducing the URLLC network slice [6], [7].
The end to end (E2E) network slicing is a very impor-
tant use-case to fulfill the requirements for service-oriented
5G networks. Many industrial organizations and standard-
ization bodies related to network communications such as
third-generation partnership project (3GPP), Fifth generation
partnership project (5GPPP), international telecommunication
union (ITU) and next-generation mobile Networks (NGMN)
are also outlined the E2E network slicing for future genera-
tion networks (FGN) [8], [9]. ETSI provides an open-source
orchestrator for slicing the network which has been developed
based on NFV specification named OSM [10]. The other open-
source solution is ONAP [11] which is developed by the Linux
Foundation to provides E2E network orchestration of virtual
appliances. Moreover, Cloudify [12] and OpenBaton [13] are
also open-source solutions for network orchestration. Although
the Cloudify was not worked on the basis of NFV MANO.
JOX [14] is another open-source orchestrator developed by
EURECOM and MOSAIC5G, it is an event-driven orches-
trator for network slicing and implemented with JujuCharms
NFs. The MOSAIC community [15] have been also provided
the solution for slicing the RAN resources with the FlexRAN
controller.
In this paper, we have proposed and developed an automated
framework named intent-based network slicing in which the
Intent-based networking (IBN) tool is used for providing the
upper-level slice configurations to the network orchestrator.
This slicing system encompasses four major modules: such
as IBN tool, Network orchestrator open-source Mano (OSM),
RAN controller FLEXRAN, and machine learning module.
This system can perform the E2E slicing mechanism for both
the core network and access network. For the 5G network,
its requirement to have an automated management system to
create, delete, and update the on-demand network slices by
providing just abstract level configurations. So, our system can
automate the procedure of network slice creation with the help
of the IBN tool. IBN tool has the ability to take the higher-
level configurations for a slice and generate the network slice
template according to network orchestrator acceptable format
&RS\ULJKW.,&6 $31206
Authorized licensed use limited to: Jeju Nat'l University. Downloaded on November 06,2020 at 02:10:04 UTC from IEEE Xplore. Restrictions apply.

for example JSON string for OSM orchestrator and JSON file
for RAN controller FlexRAN. Hence, our IBNSlicing system
can automate the slice creation process completely.
The rest of the paper is organized as follows. In Section II,
the design and architecture of E2E network slicing system is
briefly explained. The components of our IBNSclicing system
is well explained in section III. Section IV, explains the
implementation of the proposed system and its working. The
performance results generated by our proposed system are
discussed in section V. Final section contains the conclusion
of the E2E IBNSlicing system.
II. DESIGN AND ARCHITECTURE OF IBN BASED
NETWORK SLICING SYSTEM
The IBNSlicing system consists of four major modules
named as IBN tool, OSM network orchestrator, FlexRAN con-
troller, and deep learning model. This system can perform the
E2E slicing mechanism for both the core network and access
network. Moreover, our first module IBN is an automated tool
to deploy higher-level network configurations in deployment
as well as execution mode. It enables the network service
providers to configure the network resources by just providing
higher-level network configurations for both the core network
and access network. It also contains a hybrid deep learning
model GAN that continuously monitors the network resources
statistics and stores them to the IBN database resource repos-
itory. Another module is the OSM framework which provides
core network slicing support. The OSM policy configurator
of the IBN tool can convert the Higher-level configurations
into JSON string format because the OSM orchestrator accepts
the slice configurations in the form of JSON string. The
core NFVs are deployed by providing the information to the
OSM orchestrator. After that OSM NFVO dispatch, those slice
configurations to OpenStack for the deployment of network
functions (NFs). The core network slicing is managed by
the OSM framework where we deployed three separate core
network EPC functions such as MME, HSS, and SPGW for
each slice. On the other hand, the FlexRAN controller which
is SDN based controller that is used for managing and creating
the access network slicing. We develop a specific RAN slicing
policy configurator for access network which task is to convert
the higher level slice configuration provided at the IBN tool
to JSON slice template format and sent it to the underlying
FlexRAN controller. Further FlexRAN controller deploys these
configurations at eNodeB for slice creation. The complete E2E
design and architecture of IBNSlicing sysetm is presented in
Figure 1.
III. COMPONENTS OF NETWORK SLICING SYSTEM
our proposed system has four major modules: Intent-based
Networking Tool, OSM, SDN controller FLEXRAN and deep
learning (DL) module. This section contains the detail of each
module and how these modules work together and performed
Radio and Core network slicing efficiently.
A. Intent based Networking (IBN) Tool
The IBN tool encompasses six major components named
Intent Manager, Policy Store, Contract Design, Resource Man-
ager, OSM Policy Configurator, RAN policy Configurator,
and DL Module. The IBN system can able to automate
the networking slicing procedure by just providing higher-
level configurations from the front-end of the IBN tool. The
functionality of each component is as follows; the contract
design provides an easy way to users (operators, subscribers)
of the system to define their intention by using the front-end
of the IBN tool. Users of the system define their resource
requirements for creating a slice by just drags and drops at
the front-end portal of the system. The intent manager acts
as a coordinator for all the components and communicates
directly with them. When a contract comes from the operator,
it fetches the architecture information from the policy store
and generates a graph by mapping both contract information
and architecture information; after that sends those graph
information to policy configurator; where the RAN related
requirements can be sent to RAN policy configurator and
Core to OSM Core policy configurator as well. Moreover, the
network functions Information, version information, IP address
scheme, and network instances images are stored in the policy
store database repository. The Policy store is a well-organized
database repository that has all the information related to Core
network Functions as well as registered eNodeB information.
It stores the information according to the 5G architecture de-
sign. Further, the policy configurator extracts the information
related to resources from the graph which is provided by the
intent manager and converts that information in the form of a
slice template. That slice template is according to the network
orchestrator acceptable format for example for OSM Core
policy configurator can give the slice template in the form
of the JSON file which contains the information related to
the deployment of the network functions and their mapping.
The design of the OSM policy configurator is well explained
in our paper [16]–[18]. On the other hand, the RAN policy
configurator also generates the policies template in the form of
JSON which could be further sent to the FLEXRAN controller
for slice creation at eNodeB. Afterward, FlexRAN deploys
those configurations to the eNB through the master controller
and agents via Southbound API. Finally, the requested network
slice can be deployed at the access network. In this way,
dedicated EPC and RAN resources could be assigned to a
requested slice using specified S-NSSAI in the slice template.
B. Open Source Mano (OSM)
The open-source management and orchestration (OSM)
platform is a project of ETSI which provides telco operators an
E2E network service orchestrator for the deployment of net-
work services automatically. OSM has been developed based
on NFV standards and facilitate the orchestrator to interact
with other components such as infrastructure managers VIM,
NFVI, and network functions (PNFs and VNFs). Moreover,
it provides the feature of on-demand creation of networks e.g
network as a service (NaaS). It also provides an interactive

Authorized licensed use limited to: Jeju Nat'l University. Downloaded on November 06,2020 at 02:10:04 UTC from IEEE Xplore. Restrictions apply.

Fig. 1. Intent-based E2E Network slicing Architecture
GUI where the client can easily deploy, manage, and monitor
the network resources. It can also support and facilitate the
clients to deploy E2E network slices according to their need
on run-time. Although the complete life cycle of network
slices can also be managed by the OSM [10]. In our network
slicing system, the IBN component OSM policy configurator
can interact with orchestrator with the help of REST API, it
generates the slice configuration in JSON string and sent them
to orchestrator for NFs deployments. After that orchestrator
deploys the required slice resources with the help of VIM
(OpenStack).
C. SDN Controller FLEXRAN for RAN Slicing
FLEXRAN is an open-source and highly programmable
Software-defined RAN (SD-RAN) platform that separates the
RAN control plane to the data plane. It provides programma-
bility support at two levels: the first one is to develop control
and management applications over the FlexRAN controller
and the second is to develop applications within the controller
for performing the automatic deployment of control functions.
It can also control the multiple distributed base station and
their coordination among each other. The FlexRAN consists
of two main components namely FlexRAN control plane and
FlexRAN agent API. The first component FlexRAN control
plane composed of a FlexRAN master controller which is
further connected to FlexRAN agents. Although, one FlexRAN
agent is for each eNodeB and also acts as a local controller
while communicating with the master controller and other
agents. The second component FlexRAN agent API is a
southbound API which separates the data plane to control
plane such as the separation of FlexRAN control plane to the
eNB data plane. On the other hand, the FlexRAN- protocol
is responsible for the two communication between agents
and master-controller of FlexRAN. The master controller can
control and manages the operation of each eNB with the
help of FlexRAN-protocol. On the top, some monitoring and
control applications communicate with the master controller
by using the northbound-API interface. The responsibility
of these applications are to modify the RAN resources by
checking the statistics and monitoring logs of the eNB in the
FlexRAN control plane [19].
The role of the FlexRAN controller in our network slicing
system is to slice the RAN by getting the network slice tem-
plate from the IBN component RAN slicing configurator. The
FLEXRAN accepts the policies in the form of a JSON file. So,
our RAN slicing configurator converts the slice request which
is generated through the IBN system. Finally, after getting
the slice configurations from the RAN policy configurator
the FlexRAN controller deploys those slices configurations to
eNB. Furthermore, FlexRAN controller can be able to share
the RAN resources with the help of the master controller and
the agents.
D. Deep Learning Model
In this section, we are going to explain the deep learning
Model GAN and dataset used for training and testing purposes.
GAN is a very famous deep learning model that is mostly
used for image processing, video processing, and computer
vision tasks but we have used the GAN model for predicting

Authorized licensed use limited to: Jeju Nat'l University. Downloaded on November 06,2020 at 02:10:04 UTC from IEEE Xplore. Restrictions apply.

Fig. 2. FlexRAN controller abstract view
and forecasting resource utilization for the next period. It has
two modules namely generator (G) and discriminator (D):
one neural network will be used for generator and one for
the discriminator part. The first part generator generates new
data instances and the discriminator model acts like a decider
whether the data is authentic (real) or generated [20]. In our
system, we have used the LSTM (long-short term memory)
model in the Generator part and CNN (convolutional neural
network) model in the Discriminator part for the prediction of
network resource utilization (CPU, RAM, etc.).
LTSM is a very popular deep learning model mostly used
for time series data analysis, prediction, traffic forecasting,
natural language processing, and handwriting recognition. In
our case, network work traffic or resource utilization data is
time series, LTSM is the best choice to use as a generative
model to predict the slice resource utilization output based
on input dataset parameters. In the second discriminator part,
we have used the CNN model for the prediction of resource
utilization that takes the input data or generated data from
the generator module. CNN takes three-dimensional input for
training purposes and also has fully connected (FC) layers
for better prediction. This part estimates the probability of
whether the input data is real or predicted from the previous
module. Both the models (G, D) are trained on real as well as
generated datasets by using stochastic gradient descent (SGD)
model. The final output is the resource utilization statistics of
the underlying VMs for each slice next time t, after that the
predicted values will be used by the resource update manager
of the IBN tool. Those predicted statistics are very useful for
network slice acceptance, resources scale-up, failure recovery,
and resource management.
We have used an open-source dataset provided by the
MATERNA datacenter which contains performance metrics
of 547, 520, and 527 data center virtual machines (VMs).
MATERNA is a well-known service provider and IT product
Fig. 3. Design of GAN deep learning Model
manufacturing organization. The dataset was collected in three
traces with a period of three months, so one trace for one
month timestamp. The more detail about dataset collection
can be explained in this paper [21]. Although dataset has 12
attributes related to resource utilization such as of Timestamp
(ms), CPU Cores (no. of virtual Cores), CPU usage (MHz),
CPU capacity (MHz), CPU capacity (percentage), memory
requested(KBs), memory usage (KBs), memory usage (%),
received throughput(KBs), transmitted throughput(KBs), disk
writeup throughput (KBs) and Disk size (GBs). So, we trained
our model at this dataset and our deep learning model can pre-
dict the future state of the resources and update the resources
statistics IBN database. So, in this way our IBN system can
easily scale up and down the resources, in case of failure, it
will notify the IBN system and request to update the resources.
IV. EXPERIMENTAL TEST-BED IMPLEMENTATION
Our implemented test-bed consists of the IBN tool, OSM
platform, FlexRAN, OAI EPC core network functions, and
OAI eNB for achieving E2E network slicing. The other module
is the integration of machine learning model GAN which
predicts resource utilization such as CPU, RAM, etc. This
system aims to automate the network slice configuration and
management process and provide the clients with an easy
way to create network slice according to their requirements.
For testing purposes, we create three types of network slices
eMBB, IoT, and URLLC by inserting higher-level configura-
tions in the form of user intents/contracts from the IBN tool.
The IBN tool dispatches those configurations to OSM network
orchestrator and SDN-RAN controller FlexRAN for further
processing. Moreover, IBN provides higher-level contract con-
figurations in the JSON file format to OSM and FlexRAN.
Afterward, the OSM orchestrator deploys the generated
JSON string slice template configurations to the physical layer
with the help of the VIM manager OpenStack. On the other
hand, FlexRAN enforces the slice template to underlying eNBs
to allocate specific spectrum and resource blocks with the help
of the master controller and local agents. The slice template
generated through the RAN slicing configurator also provides
the configurations (IP addresses of EPC functions) for the
connectivity of dedicated EPC with selected eNB. In this way,
a specific slice is created and stitches together with Core and

Authorized licensed use limited to: Jeju Nat'l University. Downloaded on November 06,2020 at 02:10:04 UTC from IEEE Xplore. Restrictions apply.

TABLE I
TEST-BED COMPONENTS CONFIGURATIONS AND SPECIFICATIONS
Component System Specifications
FlexRAN
OS: UBUNTU 16.04
RAM: 16 GB
CPU: CORE-i5 3.0 GHZ
SSD: 500 GB
OSM
OS: UBUNTU 18 LTS
RAM: 252 GB
CPU: 32 Cores 2.10 GHZ
H/D: 2 TB
OSM Version: 7
Openstack Version: stein
IBN tool
OS: Window 10
RAM: 16 GB
CPU: Core I5 3.0 GHZ
H/D: 1 TB
Programming languages: PHP , JAVA
Database: MYSQL
SDR USRP B210 Frequency Range: 70 MHz-6GHz
Channels: 2TX * 2RX
access network automatically. The whole slice creation process
is automated where the user just used IBN GUI and define his
contract for the slice and IBN automatically deploys those
configurations to the physical layer with the help of OSM
and FlexRAN. Furthermore, we have used the EPC, eNodeB
provided by the well-known open source community OpenAir-
Interface (OAI) [22]. Further, two eNodeBs are deployed with
the help of SDR USRP B210. Three Core networks have been
deployed on different VMs by using VIM OpenStack for the
creation of three slices. Smartphones and OAI simulated UEs
are also used for testing and validation purpose. The test-bed
components with their configurations and specifications are
illustrated in table 1.
V. R ESULTS AND DISCUSSION
For our experimental setup, we have created three slices
with different QoS requirements namely eMBB, IoT, and
URLLC slices. The QoS requirements for slice 1 in user
contract is specified as 40 MB/s downlink speed and 20
MB/s uplink, slice 2 is 40 MB/s downlink and 20 MB/s
uplink, and slice 3 is 20 MB/s downlink and 10 MB/s uplink
respectively. we have performed the Iperf test to verify the
stability of each slice separately. Further, the simulated OAI
UEs were also used to test the performance of each slice. IBN
tool has the GUI, where the users can define slice contracts
by just providing higher-level configurations. Furthermore,
the specific contract is dispatched to OSM orchestrator and
FlexRAN for the deployment of network slice dynamically.
Finally, the specified uplink and downlink are reserved for
that slice with required Core EPC functions (vHSS, vMME,
vSPGW).
Figure 4 shows the downlink throughput of all three slices,
we have achieved a maximum of 31 MB/s throughput for
eMBB, 26 MB/s for IoT, and 14 MB/s for URLLC slice
downlink throughput while performing the Iperf test for each
slice. For the downlink case, we have already specified 40,
40, and 20 MB/s of throughput in slice contract for each
Fig. 4. Iperf downlink throughput for three slices.
slice. However, the configurations provided from the IBN tool
were properly deployed and configured the resources for each
slice. The uplink throughput speed tested with Iperf for each
Fig. 5. Iperf uplink throughput for three slices
slice is presented in Figure 5, as mentioned above the up-
rate throughput specified in the network slice template was
20, 20, and 10 MB/s respectively. Although we have got a
maximum of 19 MB/s for slice1 and 17 MB/s for slice2 and
7 MB/s for slice3. In our experimental setup, we have 100
MB/s connection and SDR USRP B210 also provides up to
100 MB/s spectrum support. The results show the performance
of our system is stable and the required slices were created
dynamically.
Figure 6 presents the downlink and uplink throughput test
performed with Iperf only at access network eNB. We just
provides the configurations to the RAN slicing configurator
with different QOS such as radio resource block (RBs) and
spectrum 15, 30, 45, 60,75, 90 and, 100% resources of the
RAN eNB. You can see that by increasing the resources
of RAN can increase the uplink and downlink throughput
of the network slice. when we have created a slice with

Authorized licensed use limited to: Jeju Nat'l University. Downloaded on November 06,2020 at 02:10:04 UTC from IEEE Xplore. Restrictions apply.

Fig. 6. Iperf downlink and uplink throughput test for eNB resource sharing
15% resources of eNB, it provides low 5 MB/s and 8 MB/s
throughput speed for UL and DL respectively. on the other
side, by using the 100% resources of RAN, it shows 70 MB/s
of DL and 45 MB/s of UL speed. So, after getting the slice
template configurations from the RAN slicing configurator,
FlexRAN can deploy those configurations at physical eNB and
share the resources accordingly.
VI. CONCLUSION
This paper describes the design and implementation of the
intent-based E2E network slicing platform that facilitate the
network operators to deploy the network services in a flexible
and customizable manner. The network operators instantiate
the network slice by using the GUI of IBN application where
users just input the higher-level configurations in the form of
QoS requirements. IBN system can designs the slice template
with the help of policy configurators (OSM core policy config-
urator, RAN slicing policy configurators) and dispatches those
configurations to OSM network orchestrator and FlexRAN
controller for the deployment of resources. More precisely,
our system allows network operators to automate the network
configuration and slice creation process. Moreover, deep learn-
ing model forecasts and predicts the network resources state
on runtime which helps the intent manger in deciding slice
admission. We have performed multiple tests to check the
working of our system, which show promising results for the
creation and management of the network slices. In the future,
we intend to extend our system by adding more features related
to the complete life cycle management of network slicing.
ACKNOWLEDGMENT
This research was supported by the MSIT(Ministry of
Science and ICT), Korea, under the ITRC(Information Tech-
nology Research Center) support program(IITP-2020-2017-
0-01633) supervised by the IITP(Institute for Information
communications Technology Planning Evaluation). This re-
search was one of KOREN projects supported by National
Information Society Agency (No.1711117098).
REFERENCES
[1] K. Katsalis, N. Nikaein, E. Schiller, A. Ksentini, and T. Braun, “Net-
work slices toward 5g communications: Slicing the lte network,” IEEE
Communications Magazine, vol. 55, no. 8, pp. 146–154, 2017.
[2] I. Afolabi, T. Taleb, P. A. Frangoudis, M. Bagaa, and A. Ksentini,
“Network slicing-based customization of 5g mobile services,” IEEE
Network, vol. 33, no. 5, pp. 134–141, 2019.
[3] P. W. Khan, K. Abbas, H. Shaiba, A. Muthanna, A. Abuarqoub, and
M. Khayyat, “Energy efficient computation offloading mechanism in
multi-server mobile edge computing—an integer linear optimization
approach,” Electronics, vol. 9, no. 6, p. 1010, 2020.
[4] K. Abbas, M. Afaq, T. Ahmed Khan, A. Rafiq, J. Iqbal, I. Ul Islam,
and W.-C. Song, “An efficient sdn-based lte-wifi spectrum aggregation
system for heterogeneous 5g networks,” Transactions on Emerging
Telecommunications Technologies, p. e3943, 2020.
[5] K. Abbas, K. T. Ahmed, A. Rafiq, W.-C. Song, and S.-J. Seok, “An lte-
wifi spectrum aggregation system for 5g network: A testbed,” in 2020
International Conference on Information Networking (ICOIN). IEEE,
2020, pp. 753–755.
[6] X. Li, R. Ni, J. Chen, Y. Lyu, Z. Rong, and R. Du, “End-to-end network
slicing in radio access network, transport network and core network
domains,” IEEE Access, vol. 8, pp. 29525–29 537, 2020.
[7] T. A. Khan, A. Mehmood, J. J. D. Ravera, A. Muhammad, K. Abbas, and
W.-C. Song, “Intent-based orchestration of network slices and resource
assurance using machine learning,” in NOMS 2020-2020 IEEE/IFIP
Network Operations and Management Symposium. IEEE, 2020, pp.
1–2.
[8] F. Meneses, M. Fernandes, D. Corujo, and R. L. Aguiar, “Slimano: an
expandable framework for the management and orchestration of end-
to-end network slices,” in 2019 IEEE 8th International Conference on
Cloud Networking (CloudNet). IEEE, 2019, pp. 1–6.
[9] M. Saqib, F. Z. Khan, M. Ahmed, and R. M. Mehmood, “A critical
review on security approaches to software-defined wireless sensor net-
working,” International Journal of Distributed Sensor Networks, vol. 15,
no. 12, p. 1550147719889906, 2019.
[10] OSM, “Open source mano,” https://osm-
download.etsi.org/ftp/Documentation/201902-osm-scope-white-
paper/!02-osm-scope-and-functionality.md; accessed: 1.6.2020.
[11] ONAP, “Onap: open networking automation platform,”
https://www.onap.org/; accessed: 20.5.2020.
[12] Cloudify, “Cloudify: A open source network orchestrator,”
https://cloudify.co/; accessed: 3.6.2020.
[13] OpenBaton, “Openbaton: Nfv mano-based framework,”
https://openbaton.github.io/; accessed: 25.5.2020.
[14] K. Katsalis, N. Nikaein, and A. Huang, “Jox: An event-driven orches-
trator for 5g network slicing,” in NOMS 2018-2018 IEEE/IFIP Network
Operations and Management Symposium. IEEE, 2018, pp. 1–9.
[15] Mosaic5G, “Network sharing using flexran,”
https://gitlab.eurecom.fr/mosaic5g/mosaic5g/-/wikis/tutorials/slicing;
accessed: 5.5.2020.
[16] A. Rafiq, A. Mehmood, T. Ahmed Khan, K. Abbas, M. Afaq, and
S. Wang Cheol, “Intent-based end-to-end network service orchestration
system for multi-platforms,” Sustainability, vol. 12, no. 7, p. 2782, 2020.
[17] T. A. Khan, A. Mehmood, J. J. D. Rivera, and W.-C. Song, “Machine
learning approach for automatic configuration and management of 5g
platforms,” in 2019 20th Asia-Pacific Network Operations and Manage-
ment Symposium (APNOMS). IEEE, 2019, pp. 1–6.
[18] T. A. Khan, M. Afaq, K. Abbas, A. Rafiq, A. Mehmood, and W.-C.
Song, “Generic intent-based networking approach for end-to-end slice
orchestration and lifecycle management,” , pp. 468–469, 2020.
[19] X. Foukas, N. Nikaein, M. M. Kassem, M. K. Marina, and K. Konto-
vasilis, “Flexran: A flexible and programmable platform for software-
defined radio access networks,” in Proceedings of the 12th International
on Conference on emerging Networking EXperiments and Technologies,
2016, pp. 427–441.
[20] Y. Hua, R. Li, Z. Zhao, X. Chen, and H. Zhang, “Gan-powered
deep distributional reinforcement learning for resource management in
network slicing,” IEEE Journal on Selected Areas in Communications,
2019.
[21] Materna, “Dataset,” http://gwa.ewi.tudelft.nl/datasets/gwa-t-13-materna/;
accessed: 3.5.2020.
[22] OpenAirInterface, “Oai: An open-source community,”
https://www.openairinterface.org/; accessed: 10.5.2020.

Authorized licensed use limited to: Jeju Nat'l University. Downloaded on November 06,2020 at 02:10:04 UTC from IEEE Xplore. Restrictions apply.