ADDIS ABABA SCIENCE AND TECHNOLOGY UNIVERSITY

DEPARTMENT OF ELECTRICAL AND COMPUTER ENGINEERING

(COMMUNICATION STREAM)
COLLEGE OF ELECTRICAL AND MECHANICAL ENGINEERING
ADVANCED COMPUTER NETWORK (ECEg7137)

A review on
NETWORK FUNCTION VIRTUALIZATION
By
MICHAEL TESFAYE (GSR230/12)

Submitted to Professor Cheoul-shin kang

Submitted date March 8, 2021
 Abstract
Network Functions Virtualization has emerged as an initiative from the telecom industry
(manufacturers, network operators, and carriers) to increase the flexibility of network services
deployment and integration within operator’s networks. This is achieved through the
implementation of network functions such as firewalls, encryption, filtering, load balancing,
among others, via software modules. Service function chaining (SFC) is an enabler for NFV,
providing a flexible and economical alternative to today’s static environment for Cloud Service
providers (CSPs), Application Service Providers (ASPs), and Internet Service Providers (ISPs).
The resources and networks coordinate by NFV orchestration to set up cloud-based service and
application. In this paper, a comprehensive review on the motivations and state of the art efforts
towards implementing the NFV and SFC is provided. In particular, the paper first presents the
main concepts of these new emerging technologies; then discusses various stages of SFC,
including the description, composition, placement, and scheduling of service chains. Afterward,
it assesses recent advances in this technology and points out future research directions.

Keywords: Network Function Virtualization, Service Function Chaining, Virtual Network

Functions

i
 Table of Contents
Abstract.............................................................................................................................................i

List of Acronym.............................................................................................................................iii

1. Introduction..............................................................................................................................1

1.1 Standard Architecture of NFV...............................................................................................2

1.2 Service Function Chaining....................................................................................................3

1.3 Network Slicing in NFV........................................................................................................3

2. Related Works..........................................................................................................................4

3. Conclusion................................................................................................................................7

4. Future Work.............................................................................................................................7

References........................................................................................................................................8

ii
 List of Acronym

AC Availability aware Cluster

CPU Central Processing Unit
GSPE Greedy Shortest Path Heuristic
ILP Integer Linear Programing
IoT Internet of Things
MANO Management and Orchestration
NFV Network Function Virtualization
NP Nonlinear Programming
QoS Quality of Service
SFC Service Function Chaining
SLA Service Level Agreement
VM Virtual Machine
VNF Virtual Network Function
VNF - C Virtual Network Function Composition
VNF - D Virtual Network Function Description
VNF – P Virtual Network Function Placement
VNF - S Virtual Network Function Scheduling

iii
 1. Introduction
In recent years, there is an exponential growth in the user demands for new diverse, high quality,
and agile services due to the explosion of mobile devices and the emergence of new networking
technologies such as the Internet of things (IoT) and big data networking. In the
telecommunications industry, current underlying networks are static and without any auto-
configuration abilities. In this rigid environment, service provisioning is by ordered chaining of
proprietary expensive middle-boxes deployed for each service function. On the other hand, there
is high competition between service providers in decreasing the prices and offering new diverse
services. As a result, they must find new ways to provide more agile high-quality low-price
services by reducing the complexity and cost[1].

The concept of Network Functions Virtualization (NFV) is originated to speed up the

deployment of new network services in the telecommunication domain. NFV’s main objective is
to support its revenue and future growth objectives. If we look at the present-day
telecommunication environment, it is overpopulated with a large variety of proprietary hardware
devices. If a new network service has to be launched, it requires the introduction of another
hardware entity due to which space and power complexities arise leading to much more
difficulty. Network Functions Virtualization (NFV) is a modern method of designing, deploying,
and managing network services by disassociating the physical network equipment from the
services or the functions that are running on them. NFV replaces the hardware-centric devices
with software running CPUs that work on standard servers [2].

There are many challenges faced by network operators which include a very short device life-
cycle, vendor-specific interfaces, slow protocol standardization, delay in introducing new
networking features, and many more. Networks operators were subjected to an increase in
investment needed to build hardware and cost of operation and maintenance. Due to this reason,
the network operators observed that because of network capabilities, its maintenance would be a
difficult task in the long run. Hence to deal with this difficulty and practical requirements of
space and capacity, NFV technology has been a best-seen option.

1
The main objective of NFV is virtualizing the network functions like proxies, load balancers,
firewalls, routers, or any other network function running in fixed hardware and moving them
onto virtual machines (VMs). These VMs or virtual network functions can run in servers, and all
hardware resources such as computing, storage, and networking devices are monitored as a
common resource pool. An example of the NFV model is shown below in Fig.1.

Figure 1. NFV model

1.1 Standard Architecture of NFV
NFV Architecture consists of three separate modules: VNFs, NFV Infrastructure (NFVI), and
Management and Orchestration (MANO)[1].
 Virtual Network Functions (VNFs): A VNF is a function that is implemented using
software that runs upon NFV infrastructure. A VNF runs on one or many Virtual
Machines and each of them will be managed by a module known as Element
Management System (EMS), which looks after the creation of VM-instance,
configuration, monitoring, and its security and performance. The EMS gives basic
information essential for the Operations Support System (OSS) and performs
management functionalities for the VNF-instances.
 NFV Infrastructure (NFVI): This is a virtualization layer that consists of software
resources where the VNF instances are executed. The NFVI is carried out as a set of
NFV-nodes installed in several NFVI PoPs (point-of-presence) to support different use
cases for better locality and latency. The virtualization layer is a vital element in this
NFVI domain, as it ensures lifecycles for the VNF-instances, independent of hardware.
 Management and Orchestration (MANO): MANO includes the management and
orchestration functionality required to manage the lifecycle of the VNF-instances present
on the top layer to the NFVI. The NFV-MANO communicates with the external module

2
 known as OSS/BSS, which allows it to be integrated into a well-equipped network-wide
management landscape.
1.2 Service Function Chaining
Service Function Chaining is a network capability that provides support for application-driven-
networking through the ordered interconnection of service functions. The lifecycle management
of service function chains is enabled by two recently emerged technologies, software-defined
networking and network function virtualization, that promise efficiency, effectiveness, and
flexibility gains. The implementation of SFC in a network is carried out in four stages including
Service Chain Description (SC-D), Service Chain Composition (SC-C), VNF-FG Placement
(VNF-P), and VNF Scheduling (VNF-S) [1]. The four stages are needed in practice for
delivering a service function chain optimally,
 Service chain Description (SC-D): how to describe the functional and non-functional
properties of a service (function). A service description defines what a service provides
and how it is used.
 Service chain Composition (SC-C): In the context of NFV, a network service is defined
as an entity that is composed of an ordered list of VNFs. SC-C is how to concatenate the
different VNFs efficiently to compose a network service (VNF-FG) in the most
appropriate way, concerning the requester goals.
 VNF-FG Placement (VNF-P): For a set of requested network services, VNF-P aims to
place the VNFs on suitable nodes in the network about the given objectives.
 VNF Scheduling (VNF-S): Scheduling techniques allow VNFs to share the resources to
minimize the total execution time of the network services. VNF scheduling is still a
concept under investigation; therefore, little research has been done so far.
1.3 Network Slicing in NFV
Network slicing is having a common physical infrastructure to run multiple logical networks.
That is a logical representation of the network for users between a set of network devices and
back-end applications. The virtualization and automation of the Network slicing lifecycle
management reduce the time for new slice deployment and provides tight loop monitoring and
self-adaptation to meet Service Level Agreement (SLA). On the same administrative domain,
multiple network slices may be organized jointly by administrative domain management. Some
network slices may have particular requirements like end-to-end low latency which need to be

3
taken care of throughout the process including definition, deployment, and operation of the
network slice[3].

2. Related Works
In this portion of the paper, some related articles on NFV, SFC, and Network Slicing will have
discussed.
Hassan Hawilo et al.[4] Propose an intelligent orchestrator that selects the best placement for the
VNFs in a given NFV application to minimize the intra-communication delays between the VNF
instances and enhance the QoS of the computational path (SFC). The optimized placement
achieves a higher number of VNF instances participating in a service chain with different serving
components. This outcome generates more active redundant computational paths that can be
optimally used to achieve the desired QoS in terms of performance and high availability of
service chains per request and minimize the end-to-end delay of the SFC.

Gang Sun et al.[5] analyze the power consumption in substrate networks and formulate the
problem of orchestrating an online SFC request in an energy-aware fashion across multiple
domains as an integer linear programming (ILP) model. The model’s objective was to minimize
power consumption and provide the constraints that must be met while mapping an SFC, such as
resource constraints, VNF order constraints, etc. Because the problem of orchestrating SFC
requests across multiple domains is NP-hard, we also propose a low-complexity heuristic
algorithm named energy-efficient online SFC request orchestration across multiple domains (EE-
SFCO-MD) to quickly and near-optimally solve the above model.

Adel Nadjaran Toosi et al.[6] Proposed Elastic SFC, a novel heuristic algorithm for end-to-end
latency aware dynamic auto-scaling of service function chains using horizontal and vertical
scaling of VNFs and dynamic bandwidth allocation. In dynamic bandwidth allocation, dynamic
flow scheduling is used and VNF migration to enable efficient utilization of network resources.
Evaluation of the ElasticSFC using realistic network policies and workload traces of a web
application in extended CloudSimSDN simulator to support SFC and NFV. The experimental
results showed that the proposed method significantly reduces the cost of SFC deployment.

4
Pegah Torkamandi et al. [7] propose a novel Availability-aware Clustered SFC (AC-SFC)
framework for embedding SFCs in the NFV-enabled network infrastructure which includes the
physical machines with the heterogeneous failure rate. AC-SFC uses the SPC to share backup
resources between adjacent VNFs to decrease the footprint of backup resources, network
resource consumption, and the loss ratio. Further, to overcome the complexity of the
optimization model a greedy shortest-path-based heuristic (GSPE-SFC) is proposed.

Junxiao Wang et al. [8] formulate the problem of performance and resource-aware SFC
orchestration with an optimization model Given a set of service chains in which each consists of
some network functions, we propose an approximation optimization algorithm to achieve the
guarantees of deadline and packet rate while avoiding resource idleness. To evaluate the
algorithm, prototype PRSFC-IoT upon OpenStack for online SFC orchestration is used. The
evaluation results show that PRSFC-IoT has significant advantages in deadline satisfying,
computation resource-effective utilization, deployment acceptance rate, and delay guarantee,
compared to the baseline strategy.

Guanglei Li et al. [9] implement a context-aware SFC prototype based on the hierarchical multi-
domain network scenario, and a metadata allocation scheme of Network Service Header context
headers for local and global context information sharing, which takes user/device location,
application layer protocol types, and security context into consideration. To alleviate the
bandwidth over-provision problem of context-aware SFC with function scalability, we propose a
heuristic consolidation-based SF placement algorithm proposed and an online feedback
algorithm for context-aware SFC orchestration since the online placement and chaining problem
is NP-hard. We evaluate the proposed algorithm and confirm its efficiency in a hierarchical
cloud-based edge network topology and the USA nation-wide network topology. Simulation
results confirm that compared with the naive greedy and online tabu search algorithm, the
proposed algorithm is efficient to improve the revenue-cost ratio of the SFC provision system by
compromising on node load balance.

5
Asma Chiha et al. [10] propose a cost allocation model for network slicing that aims to allocate
the cost of the network to the different network slices deployed on this network, as well as a cost
model to derive how much can be saved by using virtualization on the core network. The
proposed model requires a slice dimensioning model to distribute the cost of the network to the
running slices based on their consumption in terms of network resources. The resource allocation
model or slice-dimensioning model demands the hardware requirements of each 5G network
functions NFs as inputs.

Xi Huang et al. [11] proposes a novel model that separates the granularity of system state
characterization and strategy making. In particular queuing model at the requested granularity to
characterize system dynamics was developed. Unlike flow-level abstraction, this model requires
no prior knowledge on underlying flows but accurately captures the interplay between successive
instances. To enable online and efficient decision-making, the long-term stochastic optimization
problem transformed into a series of sub-problems over time slots. The paper address the
dynamic service chaining and scheduling problem in the NFV system by jointly considering
resource utilization, energy efficiency, and request latency. The results show that prediction with
mild false-positive conduces to shorter response times.

Kaimig Liu et al. [12] establish a mathematical model with multi-parameter constraints firstly.
When defining the weight of routing cost, the weight factors are adjusted for different types of
flows, to achieve different routing paths for different characteristic flows and strictly guarantee
the flow's QoS requirements. To avoid the solution trapped in local optimality and decrease the
complexity, an improved routing algorithm based on Viterbi is adopted, which eliminates some
function nodes who occupy too many resources according to the network resource occupancy, it
can reduce the selection solution space to reduces the complexity of the algorithm effectively.
The simulation result shows the proposed algorithm can improve the SFC flows reception rate,
moderately reduce forwarding hops and balance traffic load compared with the existing
algorithms.

6
 3. Conclusion
In this paper, we present a comprehensive review of NFV which includes three big parts, that is,
the basic concepts of NFV (including motivation, terminologies, standardization efforts, history,
and architecture), VNF related algorithms (including VNF placement, scheduling, migration,
chaining, and multicast), challenges and future directions. In particular, the NFV architecture is
introduced using a bottom-up approach to highlight its hierarchical structure. This review
introduces the main concepts and background of two new emerging technologies in networking:
Network function virtualization and Service function chaining. Some main papers in these two
fields are reviewed and various approaches and techniques are discussed.

4. Future Work
Even though many researches are done on NFV, it is still in the infancy. Therefore, researchers
will continuously focus on the extensive activity around NFV soon as there are so many new
topics to be explored, for instance, the migration path to NFV, NFV inter-operability, and service
composition. There are great opportunities for future work in SCF. The high resiliency
requirement of Service Function Chaining is one of the main features, which should be paid high
attention to during the deployment phase.

7
 References
[1] G. Mirjalily and Z. Luo, “Optimal Network Function Virtualization and Service Function

Chaining: A Survey,” Chinese J. Elect., vol. 27, no. 4, pp. 1-14, 2018.

[2] B. Yi, X. Wang, K. Li, and M. Huang, “Review article A comprehensive survey of

Network Function Virtualization,” vol. 133, pp. 212–262, 2018.

[3] T. Lin and Z. Zhou, “NFV-Enabled Network Slicing,” 2018 IEEE Int. Conf. Commun.

Work. ICC Work. 2018 - Proc., pp. 1–6, 2018.

[4] H. Hawilo, M. Jammal, A. Shami, and S. Member, “Network Function Virtualization-

Aware Orchestrator for Service Function Chaining Placement in the Cloud,” IEEE J. Sel.

Areas Commun., vol. 37, no. 3, pp. 43–55, 2019.

[5] G. Sun, Y. Li, H. Yu, A. V Vasilakos, and X. Du, “Energy-efficient and traffic-aware

service function chaining orchestration in multi-domain networks,” Futur. Gener.

Comput. Syst., vol. 91, pp. 347–360, 2019.

[6] Toosi, A.D.; Mahmud, R.; Chi, Q.; Buyya, R. Management and Orchestration of Network

Slices in 5G, Fog, Edge and Clouds. Fog and Edge Computing: Principles and Paradigm;

John Wiley and sons: Hoboken, NJ, USA; pp. 79–101, 2019.

[7] P. Torkamandi, “Availability Aware SFC Embedding in NFV : A Clustering Approach,”

IEEE J. Sel. Areas Commun., vol. 38, no. 3, pp. 1-12, 2019.

[8] J. Wang, H. Qi, K. Li, and X. Zhou, “PRSFC-IoT: A performance and resource aware

orchestration system of service function chaining for internet of things,” IEEE Internet

Things J., vol. 5, no. 3, pp. 1–10, 2018.

8
[9] G. Li, H. Zhou, B. Feng, and G. Li, ‘‘Context-aware service function chaining and its

cost-effective orchestration in multi-domain networks,’’ IEEE Access, vol. 6, pp. 76–91,

Jun. 2018.

[10] A. Chiha, M. Van der Wee, D. Colle, and S. Verbrugge, Network Slicing Cost Allocation

Model, vol. 28, no. 3. Springer US, 2020.

[11] X. Huang et al., “Online VNF Chaining and Predictive Scheduling : arXiv, vol. 19, no. 4,

pp. 1–36, 2019.

[12] K. Liu and X. Li, “An Improved Routing Algorithm for Service Function Chain in

SDN/NFV Networks,” 2019 IEEE 5th Int. Conf. Comput. Commun. ICCC 2019, pp. 43–

47, 2019.