Total Cost Modeling for VNF based on Licenses and Resources
Ghoshana Bista
1,2
, Eddy Caron
1 a
and Anne-Lucie Vion
2
1
UMR CNRS - ENS Lyon, UCB Lyon 1 - Inria 5668, Lyon, France
2
Orange, Pessac, France
Keywords:
License Cost, Simultaneous Active Users, Software Licensing, VNF, Software Cost, Cloudification, Soft-
warization.
Abstract:
Moving to NFV (Network Function Virtualization) and SDN (Software Defined Network), Telco cloud archi-
tectures face four key challenges: interoperability, automation, reliability, and adaptability. All these chal-
lenges involve the optimization of resources; whether it is to increase the utilization of hardware resources
(virtualization) or to deliver shared computing resources and functions in real-time (cloudification). Soft-
warization of networks is a consequence of telecom cloudification. Virtual Network Function (VNF) is pro-
tected by IPR (Intellectual Property Right) like any software, ensured by a license describing usage rights and
restrictions at a given cost. Until now limited studies have happened in the economic dimension linked to
softwarisation. Currently, the telco industry struggles to converge and standardize licensing and cost models.
At risk: the network cloudification benefits could be swept away by poor management of resources (Hardware
and Software). This article presents a preliminary model for optimizing the total cost of a VNF, based on the
Resource Cost (RC) and License Cost (LC). This analysis is inspired by measurement and licensing practices
commonly observed in the Telcos industries,i.e consumption and capacity.
1 INTRODUCTION
NFV aims to increase automation and network reli-
ability for better and quicker service delivery. From
the research and markets, the global NFV market is
projected to grow from 12.9bn dollars in 2019 to
36.3bn dollars by 2024
1
. Also, 60 percent of ser-
vice providers will adopt NFV in the next two year
1
.
Service providers, more specifically telco companies,
need to adapt to this shift quickly and efficiently.
More than only technological challenges, historical
telcos must face the arrival of large hyper-scalers,
partners, and also aggressive competitors.
Service Providers will benefit from NFV if they
can enable new services with a faster time-to-market,
rapidly scaling resources up and down, lowering the
costs. The key challenges facing NFV are thus linked
with resource optimization. Success relies on the abil-
ity to monitor and use standards and interoperable re-
sources: in other words, to mix and match various
software components on standard COTS (Commer-
cial -Off-The-Shelf) hardware.
a
https://orcid.org/0000-0001-6626-3071
1
https://www.f5.com/ f r
f
r/company/blog/why-nfv-is-
more-relevant-than-ever
As the network becomes software, failure in con-
trolling software spending destroys the promises of
NFV efficiency. The paradigm shift from equipment
property toward SW (Software) Right To Use (SW
RTU) is adding complexity in resource management.
As SW is protected by IPR over a license, it becomes
essential to ensure the compliance of SW deploy-
ments regarding acquired rights. As well, it becomes
essential to optimize license costs. For this, one of the
common practices in IT, extending to Network, is to
practice Software Asset Management (SAM). Imple-
menting end-to-end SAM guarantees that users buy
all the licenses user need, only the license need: to
avoid counterfeiting and waste.
SW license frames the rights and obligations of
the CSP (Communication Service Providers) to use
SW. License is associated with a cost (LC) which de-
pends on the volume of rights granted. The volume
granted and associated conditions of uses are contrac-
tually defined by one or several metrics. There are
currently, no standards on metrics and their definition,
they depend on the creativity of software providers.
Which could facilitate every supplier of VNF to pro-
pose their metrics, model, and tools. And, this is be-
ing an intriguing and complex task for VNF services
246
Bista, G., Caron, E. and Vion, A.
Total Cost Modeling for VNF based on Licenses and Resources.
DOI: 10.5220/0011079600003200
In Proceedings of the 12th International Conference on Cloud Computing and Services Science (CLOSER 2022), pages 246-253
ISBN: 978-989-758-570-8; ISSN: 2184-5042
Copyright
c
2022 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
providers. Based on our observation on the telcos in-
dustries, metrics can be linked, among with usage, re-
source allocation, and resource consumption. Thus, it
is not wiser to consider LC and RC independently, but
we need to consider that LC can be dependent on the
resources and vice versa.
In this paper, we propose a model to evaluate and
optimize TCO (Total Cost of Ownership) of VNF de-
ployment based on LC and RC. We base our model
on two metrics; Simultaneous Active User (SAU) and
Bandwith (BW) as we observed that they are well
known in the telcos industries and can fit both re-
source consumption, allocation, and capacity
textitOur contribution:
We formulate the software cost model for VNF
based on LC, RC, and TC
We introduce the concept of license reference,
which is crucial to anticipate the LC, RC, and TC.
We provide different methods for licensing and
estimating its associated cost as LC, RC, and TC.
The rest of the paper is organized as follows: Sec-
tion 2 provides our related work and background.
Section 3 present our models and simulation and Sec-
tion 4 concludes this research.
2 BACKGROUND AND RELATED
WORK
2.1 Business Model
Our assumption of the business model is similar to
(Ahvar et al., 2017), which is defined as (i) Service
provider, (ii) Network operator, (iii) VNF provider,
and (iv) User, customer or client, end-user. For us, a
service provider (SP) is a telco company that provides
communication services, network operator provides
servers, and operates infrastructure. VNF provider
is those that provide VNF and end-user are service
users. Different entities have different roles but these
can be changed or they can be operated by the same
entities too. Total Cost (TC) is calculated based on
LC (License Cost) and Resource cost (RC). Although
resources cost occasionally includes LC and various
other costs such as link, maintenance, upgrade, hard-
ware cost. In this study, we considered only RC
as VNF instances required to operate the necessary
amount of license (license reference). License is an
agreement that comes with rights and duties. The
right is to use a certain amount of SAU or BW in the
respective VNF and duties are to comply.
2.2 Related Work
Some researches were focused on optimizing the net-
work cost and the network path most of them include
LC as a constant entity. In (Mouaci et al., 2020) they
have dealt with finding the best place of VNF for a
better routing path for each demand this article helps
our research for finding the placement of SFC (Ser-
vice Function Chain). (Ahvar et al., 2017) had a sig-
nificant impact on licensing but still, the authors did
not focus on the licensing cost or providing better op-
tions for clients. In (Kiran et al., 2020) they developed
VNFPRA problem which finds the optimal placement
of VNFs in SDN/NFV-enabled MEC (Mobile Edge
Computing) nodes to reduce the deployment and re-
source cost using genetic and mixed-integer prob-
lems. From this article, we get an idea about resource
allocation. In (Liu et al., 2019) they focused on the
placement of the VNF and traffic steering using net-
work cost, node cost, and VNF placement cost. How-
ever, this article did not discussed the total cost and
optimizing LC but has illustrated a good insight into
in-network cost. In (Pham et al., 2017), Chuan Pham
et al. formulated problems for joint optimization and
traffic cost optimization using the Markov Approxi-
mation (MA) in which they added their matching ap-
proach called SAMA. This research helps us to get
a proper idea of VNF instances, Although this work
was huge and concrete, they have not considered li-
cense cost or overall total cost. In (Shi et al., 2015)
authors focused on resources allocation of NFV com-
ponents using Markov Decision Process and Bayesian
learning which helped to dynamically allocate NFV
components.
3 FROM SYSTEM MODEL TO
SIMULATION
3.1 System Model
In this section, we provided cost models based on
VNF and traditional ways.
3.1.1 Traditional Ways
The popular traditional model is a perpetual license
and pay as you grow.
Perpetual license: In this system users have to pay
upfront then only users have the right to use the
software. Depending upon the license entitlement
user can upgrade and update their software. Since
it is one-time pay generally it is highly expensive.
Total Cost Modeling for VNF based on Licenses and Resources
247
The dimension parameters used for a perpetual li-
cense are:
LC
prl
: One time cost, upfront payment for li-
cense.
RC
prl
: One time cost , upfront payment for re-
sources.
TC
prl
: Total cost for perpetual license.
TC
prl
= LC
prl
+ RC
prl
(1)
Pay as you grow (PAYG): It is the model in which
end-user have to pay according to their capacity
increment, it can be usages based on resources,
services, or others. There are lots of pay-as-you
methods such as pay as you use, pay as you eat,
pay as you go, but for us, it is paid as you grow,
and grow here is in terms of SAU/BW. The dimen-
sion parameters used for the PAYG license model
are:
no
SAU
= Number of SAU at a time.
CS
PAY G
: Unit cost per SAU for license.
CSr
PAY G
: Unit cost per SAU for resources.
LCs
PAY G
: License cost for SAU.
RCs
PAY G
: Resource cost for SAU.
TCs
PAY G
: Total cost for SAU using PAYG.
LCs
PAY G
= Cs
PAY G
× no
SAU
RCs
PAY GS
= CSr
PAY G
× no
SAU
(2)
TCs
P
AY G = LCs
PAY G
+ RCs
PAY G
(3)
Now for the BW, dimension parameters are;
no
BW
: number of SAU at a time.
CB
PAY G
: Unit cost per BW for license.
CBr
PAY G
: Unit cost per BW for resources.
LCB
PAY G
: License cost for BW.
RCB
PAY G
: Resource cost for BW.
TCB
PAY G
: Total cost for BW using PAYG.
LCB
PAY G
= CB
PAY G
× no
SAU
RCB
PAY GS
= CSr
PAY G
× no
SAU
(4)
TCB
PAY G
= LCB
PAY G
+ RCB
PAY G
(5)
One of the important points here not to forget is
that these are actually business models, not actual li-
censing models, they are only used as license mod-
els due to lack of the standard license models and
metrics. To fill this gap we proposed license models
which ultimately help to construct optimized business
models in a virtual environment, VNF.
3.1.2 Virtual Network Function
Our crucial problem was to find reliable and authentic
methods for licensing which would ultimately help to
model optimize the total cost in VNF. To address this
task we used the two important license metrics that
are SAU and BW (Bandwidth).
SAU: SAU generally means simultaneous active
users connected with VNF who are consuming
some resources, and using services provided by
VNF.
BW: It is related to the amount of bandwidth-
consuming/consumed by SAU.
License reference: it is considered to be the es-
timated number of licenses required for the VNF
system for a certain period.
We choose these metrics among the other existing
metrics such as Transmission Per Second (TPS), Re-
quest Per Second (RPS), etc. because these metrics
are convenient to measure the usages and scalable pa-
rameters. Using these two metrics we created two
license reference (LR) models LR
SAU
(Equation (6))
and LR
BW
(Equation (7)) respectively for SAU and
bandwidth, they can be formulated as follows;
LR
SAU
= max
jD
iV
average(SAU
i
, j)
(6)
Similarly using BW,
LR
BW
= max
jD
iV
average(BW
i
, j)
(7)
where, V = (1, 2,.....v) is a set of all concerned
VNF (can be same or different type) in node. We de-
fine H as a set of hours (1, 2, 3, ..., h), D as a set of
days (1, 2, 3, ..., d) and R as a set of License Refer-
ence (LR). LR corresponds to LR
SAU
or LR
BW
which
we get from the Equations (6) and (7). Table 1 de-
lineates the parameters use in our formulation. Our
assumption for this research was that all VNFs were
deployed properly in their respective places and they
were functioning accurately in full capacity. These li-
cense references helped to generate an optimized total
cost model which includes license and resource cost
calculated as;
TC = RC
j
+ LC
j
(8)
Now, LC and RC can be calculated for capacity
model for a day, j D and r R, as:
LC
ca
j
= φ
r
+ σ
r
α
r
τ
r
, (9)
RC
ca
j
= θ
r
+ δ
r
τ
r
β
r
(10)
CLOSER 2022 - 12th International Conference on Cloud Computing and Services Science
248
LC and RC can be calculated for consumption
model for a day, j D, as:
LC
cp
j
= γ
j
× α
r
, (11)
RC
cp
j
= γ
j
× β
r
, (12)
Table 1: Parameters use in problem formation.
r A License Reference. r R,
φ
r
Pre-paid amount for License Refer-
ence (C),
τ
r
Surpass or exceed License Refer-
ence r,
α
r
Unit cost of license for License Ref-
erence r (C/SAU or Mbps),
σ
r
License factor for License Refer-
ence r,
θ Prepaid amount for resources (C),
δ
r
Resources factor for License Refer-
ence r,
β
r
Unit cost of resources for License
Reference r,
γ
j
License Reference for a day, j D,
LC
ca
License cost for capacity,
RC
ca
Resources cost for capacity,
LC
cp
License cost for consumption,
RC
cp
Resources cost for consumption.
D As a set of days (1, 2, 3, ..., d)
Now, implementing these equations (6), (7) and
(8) in different scenario, such as i) VNF instances sce-
nario ii) Users dependent iii) Using flavour in nodes.
These scenarios are depends on users, usages, and
nodes.
3.2 Simulations
In this section, we deal with two types of use case
scenarios and a real scenario.
3.2.1 Scenario 1: VNF Instances
In this scenario, we presented different available
clients’ usages like; Web, VoIP, and Online Game.
We modified the table of (Pham et al., 2017) to ad-
just it to our model. For this scenario, users need to
be aware of their requirements based on SAU or BW.
Using SAU and BW the flavour base table is proposed
on Table 2 and Table 3 which is supposed to meet the
user’s requirement. Importantly, license cost and re-
source cost mentioned in our tables are LC
ca
and RC
ca
i.e. license capacity and resources capacity cost.
So, if the traffic requirement is BW (65Mbps) for
web services from Table 4 then for this services sim-
ulator proposed flavour E and for VoIP simulator pro-
posed flavour D. Whenever exact traffic range require-
ment is not available simulator proposed higher value
Table 2: BW Flavour Table for Scenario 1 (VNF instances).
Flavour BW
(Mbps)
LC
cost
(Ke)
vStorage
(TB)
vRAM vCPU Redundancy Resources
cost (Ke)
A 15 200 2 2 GB 2 1 170
B 20 250 3 2 GB 3 2 200
C 30 355 4 3 GB 4 2 285
D 50 435 4 3 GB 4 2 338
E 65 549 5 4 GB 4 2 420
F Customize your needs
Table 3: SAU Flavour Table for Scenario 1 (VNF in-
stances).
Flavour SAU LC
cost
(Ke)
vStorage
(TB)
vRAM vCPU Redundancy Resources
cost (Ke)
A 100 250 4 2 GB 2 1 100
B 150 350 4 3 GB 3 2 150
C 200 450 6 4 GB 4 2 215
D 250 550 7 5 GB 5 2 275
E 350 650 8 5 GB 5 2 325
F Customize your needs
Table 4: Service chain of different client usages with BW.
Client usage Service Chain Minimum
Traffic Re-
quired (BW)
Minimum
Traffic Re-
quired (SAU)
Web Services NAT-FW-
WOC-IDPS
65Mbps 165
VoIP NAT-FW-
TM-FW-NAT
35Mbps 200
Online Game NAT-FW-
VOC-WOC-
IDPS
150Mbps 300
NAT: Network Address Translator, FW: Firewall, TM: Traffic Moni-
tor, WOC: WAN Optimization Controller, IDPS: Intrusion Detection
Prevention System, VOC: Video Optimization Controller
flavour, and for the last online game since users re-
quirement is higher than the available option so users
either can be satisfied with flavour D or the best op-
tion would be to customize their requirements with
services provider, i.e. flavour E. So, in these kinds
of situation simulator directly proposed flavour E. A
similar phenomenon goes for SAU as well. For web
services usage, flavour C was proposed from Table 3.
Similarly, for VoIP, flavour C matched the require-
ment too and for the online game, flavour E covered
the user’s requirement. So using these flavours based
tables the end-user can estimate the optimized total
cost.
3.2.2 Scenario 2: Users
The second scenario depends on the user. In this
research users were categorized in two types, Ê
Resources Know Users and Ë Resources Unknown
Users.
Ê Resources Known Users (RKU): These are the
users who know the estimated amount of resources
in terms of SAU/BW required for their system. In
this scenario, flavour table was created based on SAU,
BW like in previous Table 2 and Table 3, also in
these tables total cost was introduced so that depend-
Total Cost Modeling for VNF based on Licenses and Resources
249
ing upon the user’s budget they can choose flavours
too.
For this evaluation, a simulator had been cre-
ated named flavour selector where users can input the
range of SAU, BW, and total cost depending upon
their requirement our simulator will propose the ap-
proximate flavor. For example, if the SAU range
given by the client is 100-120 our selector suggests
flavor A from Table 5. Another case is a range be-
tween 130-160 then our selector suggests flavour B
also whenever the user inputs range value then it will
suggest customizing flavor, i.e. flavour E. Similar to
BW, if the client provides the range of bandwidth be-
tween 18-20 Mbps the selector will propose flavour B
from Table 6. Another interesting case is using to-
tal cost. For the total cost, we added the value of
Resources Cost (RC
ca
) and License Cost (LC
ca
) as
shown in column (9
th
) in Table 5 and Table 6. So,
depending upon the customer’s budget simulator pro-
posed flavour between SAU and BW. For example:
if the client’s budget range is from 800-900ke, the
simulator provides flavour D from Table 5. Further-
more, if the range is from 600-700ke then it could be
from flavour C from SAU or BW table. So, to avoid
the confusion of choosing between SAU and BW the
simulator asks the client preference between SAU and
BW. Thus, depending upon the client’s needs simula-
tor provides the result either from BW, SAU or cost.
Ë Resources Unknown Users (RUU): These are
the users who don’t have the estimated knowledge
about resources requirements (SAU, BW) for their
system. So for these types of users, a simulator was
created to provide the users with several choices. At
first, users need to provide their range after which the
simulator will propose the least SAU value from the
Table 5. If the user is not satisfied with that proposed
then they can processed further simulator will pro-
pose from BW Table 6, least range from BW. If this
range is also not satisfactory to the client requirement
then the simulator proposed the mean value from the
SAU flavour table, if this also fails to meet the user’s
needs the simulator proposed the mean value from
the BW flavour table. After this, the simulator pro-
posed the highest value of SAU and BW from SAU
and BW flavour table respectively. So the simulator
proposed from least to maximum flavour value from
tables based on SAU, BW. Thus, our aim here is to
provide as many options as possible to the user. Ad-
ditionally, the offer can be made concerning the total
cost as performed in RKU.
3.2.3 Scenario 3: Node Analysis in Real Scenario
For this analysis, we used the two techno-economic
friendly models to estimate the LC, RC. They are
Table 5: SAU Flavour Table for Scenario 2.
Flavour SAU LC
(Ke)
vStorage
(TB)
vRAM vCPU Redundancy Resources
cost
(Ke)
Total
cost
(Ke)
A 100 250 4 2 GB 2 1 100 350
B 150 350 4 3 GB 3 2 150 500
C 200 450 6 4 GB 4 2 215 665
D 250 550 7 5 GB 5 2 275 825
E 350 650 8 5 GB 5 2 325 975
F Customize your needs
Table 6: BW Flavour Table for Scenario 2.
Flavour BW
(Mbps)
LC
(Ke)
vStorage
(TB)
vRAM vCPU Redundancy Resources
cost (K
e)
Total
cost
(Ke)
A 15 200 2 2 GB 2 1 170 370
B 20 250 3 2 GB 3 2 200 450
C 30 355 4 3 GB 4 2 285 640
D 50 435 4 3 GB 4 2 338 773
E 65 549 5 4 GB 4 2 420 969
F Customize your needs
capacity and consumption. To adapt these models
from a business point of view we have considered
some thresholds, constraints related to license and re-
sources like license threshold, resources threshold, li-
cense factors, etc.
Capacity: The capacity analysis is similar to pre-
paid service where a certain amount of cost is
paid upfront to a certain capacity (license refer-
ence for our research) of VNF. When it surpasses
the threshold extra costs will be incurred. The
threshold can be a license or resource or both.
In this research both LC and RC were estimated
using unit cost and license reference using Equa-
tions (9) and (11). In this mode, once the capacity
is increased it cannot be reversed even if the con-
sumption (SAU/BW) is lower than the threshold.
Consumption: Clients will pay for the resources
they had consumed or will consume during a cer-
tain time. As a consequence, there is no contrac-
tual threshold limiting the user’s ability to con-
sume resources (SAU, BW). It was calculated us-
ing Equations (11) and (12).
LC threshold: This is the threshold for calculating
license cost. LC threshold was implemented for
both LC
ca
and LC
cp
. In this research, wherever li-
cense is being used it includes both (LC
ca
, LC
cp
).
It is defined at the time of negotiation of the con-
tract based on estimated needs. If the uses exceed
the threshold, the cost of a license is increased by
1.5, 2, 3, . . . , n, which is known as license factor
σ. The license threshold is based on LR.
RC threshold: It is the threshold in the resources.
Whenever a threshold is exceeded, it requires a
careful evaluation to understand whether the ex-
ceeded threshold can be covered by a single re-
source or more. If it can be covered with one re-
source then our research will use resources fac-
tor (δ)=1, and if requires more than one it will be
from 1,5, 2, 3, 4, 5, . . . , n. This threshold is also
dependent upon the LR. Alike the LC threshold, it
CLOSER 2022 - 12th International Conference on Cloud Computing and Services Science
250
can be negotiable between the VNF provider and
the SP.
LC factor (σ): It is a multiplicative factor after ex-
ceeding threshold, σ= 1.5, 2, . . . , n. License factor
and resources factor were introduced here to cre-
ate a proper business model because whenever the
threshold is exceeded in the capacity model, the
service provider will charge some extra amount.
RC factor (δ): It is similar to the LC factor but it
is in the resources aspect. δ= 1.5, 2, 3, . . . , n.
So, now using all these metrics and equations
(12) and (11) the total cost in consumption model be-
comes:
max
jD
(α
r
γ
j
+ β
r
γ
j
) (13)
Similarly, from equations (10) and (9) total cost
for capacity model became:
max
jD
(φ
r
+ σ
r
τ
r
α
r
+ θ
r
+ δ
r
τ
r
β
r
) (14)
A presumption was made that it will meet the QoS
threshold, T H
f
, i.e min TC T H
f
. T H
f
is not a nu-
merical value but a condition.
Now, for this situation, we considered a similar
scenario as in (Malandrino et al., 2019) i.e ICA (In-
tersection Collision Avoidance). ICA issues the alert
signal if any pair of them are about to collision. All
the parameters which were adapted to replicate the
business models are given in Table 7 and Table 8.
This simulation is executed on the Intel(R) Core(TM)
i7-6600U CPU @ 2.60GHz 2.81 GHz, 16GB RAM,
Windows 10. For the sake of simplicity, we as-
sume that resources can be scaled easily. We gen-
erated SAU and BW randomly on each virtual node,
also known as Virtual Evolve Packets (vEPC) such as
MME, SGW, PGW, etc. After the SAU and BW were
generated in each vEPC we implemented our license
reference model as in Equations (6) and (7) and ob-
tain the Figures 2 and 6 for SAU and BW. After the
estimation of license reference, we estimated the li-
cense cost using the Equations (9) and (11) and its
result is shown in the Figures 3 and 7. After success-
fully evaluating the LC we analyze the RC using the
Equations (10) and (12) with the help of the license
reference as shown in Figures 4 and 8. Further, we
estimated the total cost using Equations (13) and (14)
which were shown in Figures 5 and 9. The experiment
is carried out three times with three different random
values for thirty days and its cumulative average re-
sults are presented in all figures.
3.2.4 Evaluation
Figures 2 and 6 show the estimated license reference
based on randomly generated BW. Randomly gener-
Figure 1: VNF graph of the ICA service.
Table 7: Simulation parameter for SAU.
Parameters Value
License threshold for ca-
pacity
4725
Resource threshold for
capacity
4725
σ for both capacity for
SAU
1.5 (for LR less then
4725, 2 for LR >4725
δ 1.5 for SAU < 4725 and
2 for LR > 4725
θ 33000 (C)
φ 40000 (C)
SAU 500 add random(0,50),
random(0,100) and ran-
dom(10,1000)
α for SAU 10C per LR for License
β for SAU 6C per LR for Resource
CS
PAY G
0.01(C) per SAU
CSr
PAY G
0.04 (C) per SAU
LC
prl
500 (C)
RC
prl
800 (C)
Table 8: Simulation parameter for BW.
Parameters Value
License threshold for ca-
pacity
65
Resource threshold for
capacity
70
σ for both capacity for
BW
1.5 (for LR less then 65, 2
for LR >65)
δ 1.5 for SAU < 70 and 2 for
LR > 70
θ 700 (C)
φ 600 (C)
BW random(5,10),random(10,20)
and random(30,40)
α for BW 10 C per LR
β for BW 6 C per Resource
CB
PAY G
0.01 (C) per BW
CBr
PAY G
0.04 (C) per BW
LC
prl
500 (C)
RC
prl
800 (C)
ated BW is also shown in the figures with help of
which pay as you grow and perpetual license and its
related cost were estimated. We can see on both fig-
ures that daily BW is higher than the reference. It is
due to licensing reference being based on hourly av-
erage, maximum over a day from all concerned VNF
from Equations (6) and (7). Not to be confused that
daily SAU and BW shown in figures are 24 hours con-
Total Cost Modeling for VNF based on Licenses and Resources
251
Figure 2: License Reference for SAU.
Figure 3: Cumulative License cost using SAU.
Figure 4: Cumulative Resource cost using SAU.
sumption by ICA service. Figures 3 and 7 which are
the cumulative license cost for 30 days we can see
that license cost using perpetual is lower and license
cost using pay as you grow is higher. An interesting
case here is license cost using consumption and ca-
pacity methods these are lower than pay as you grow,
among consumption and capacity, consumption has a
lower cost than capacity. One can argue that since
the perpetual cost is lower why not choose it but it
is not beneficiary for the VNF services provider. Be-
cause the perpetual model did not consider the usages
or resources consumption it is not fair for the VNF
services provider. Now, coming back to our figures in
Figure 5: Cumulative Total cost using SAU.
Figure 6: License Reference for BW.
Figure 7: Cumulative License cost using BW.
contrast with LC of SAU, BW LC for consumption is
higher and capacity is lower. Figures 4 and 8 show
that the consumption model estimated the lower re-
sources cost than the capacity in both SAU and BW.
Figures 5 and 9 show the estimated total cost. Fig-
ure 5 is the total cost for the SAU here we can see
that the consumption model estimated the lower cost
than capacity. While the BW capacity model esti-
mated the lower cost than consumption as shown in
Figure 9. One of the interesting points we can depict
from these figures is that when we use SAU as met-
rics with our models’ consumption model estimated
lower cost but in contrast of this when we use BW
CLOSER 2022 - 12th International Conference on Cloud Computing and Services Science
252
Figure 8: Cumulative Resource cost using BW.
Figure 9: Cumulative Total cost using BW.
as metrics with our models capacity model estimated
lower cost. This is due to the consumption of BW and
the number of SAU can not be compared, these are
two different metrics. SAU is a user connected with
the VNF and BW is the consumption of bandwidth
based on the SAU or services it consumes. This leads
us to the point that licensing could be done in differ-
ent ways depending upon the end-user requirement,
scenario, and QoS parameters (throughput, bit-rate,
etc). Also, different licensing metrics can be consid-
ered and depending upon the licensing metrics LC,
RC, and TC cost can be optimized. Thus, in simula-
tion for ICA when we use SAU as a metric consump-
tion model generated an optimized LC, RC and when
BW as a metric capacity model provided optimized
LC and RC.
4 CONCLUSIONS
The study has tried to suggest several models that are
relevant to the various scenarios. We compare the tra-
ditional ways of estimating total cost and our models
(capacity and consumption). Results confirm that our
model is far better than the traditional one. We also
present different kinds of possible scenarios such as
VNF instances and Users which have a huge range of
requirements to be fulfilled. To meet the scenario re-
quirement, we proposed the flavours methods (SAU,
BW). For users, we presented a solution based on to-
tal cost, SAU, and BW. Secondly, we implemented
our models in real VNF scenario ICA which clearly
shows that our models outperform other traditional
models it is because we introduced the concept of
license reference based on SAU and BW. So, from
all these, we can conclude that licensing is a com-
plex task that depends not only on one factor or met-
rics but also on several metrics, users, services, and
many others. We tried to include potential metrics
and constructed a novel model. Thus, we assume that
our models are not just limited to one scenario but
could be implemented on different circumstances and
topologies and will be helpful to estimate the opti-
mized total cost. Our future work will be to enhance
this model using DF, green energy, and implement it
on the more complex VNF SFCs.
REFERENCES
Ahvar, S., Phyu, H. P., Buddhacharya, S. M., Ahvar, E.,
Crespi, N., and Glitho, R. (2017). Ccvp: Cost-
efficient centrality-based vnf placement and chaining
algorithm for network service provisioning. In 2017
IEEE Conference on Network Softwarization (Net-
Soft), pages 1–9. IEEE.
Kiran, N., Liu, X., Wang, S., and Yin, C. (2020). Vnf place-
ment and resource allocation in sdn/nfv-enabled mec
networks. In 2020 IEEE Wireless Communications
and Networking Conference Workshops (WCNCW),
pages 1–6. IEEE.
Liu, Y., Pei, J., Hong, P., and Li, D. (2019). Cost-efficient
virtual network function placement and traffic steer-
ing. In ICC 2019-2019 IEEE International Confer-
ence on Communications (ICC), pages 1–6. IEEE.
Malandrino, F., Chiasserini, C. F., Einziger, G., and Scalo-
sub, G. (2019). Reducing service deployment cost
through vnf sharing. IEEE/ACM Transactions on Net-
working, 27(6):2363–2376.
Mouaci, A., Gourdin,
´
E., Ljubi
´
C, I., and Perrot, N.
(2020). Virtual network functions placement and
routing problem: Path formulation. In 2020 IFIP
Networking Conference (Networking), pages 55–63.
IEEE.
Pham, C., Tran, N. H., Ren, S., Saad, W., and Hong, C. S.
(2017). Traffic-aware and energy-efficient vnf place-
ment for service chaining: Joint sampling and match-
ing approach. IEEE Transactions on Services Com-
puting, 13(1):172–185.
Shi, R., Zhang, J., Chu, W., Bao, Q., Jin, X., Gong, C., Zhu,
Q., Yu, C., and Rosenberg, S. (2015). Mdp and ma-
chine learning-based cost-optimization of dynamic re-
source allocation for network function virtualization.
In 2015 IEEE International Conference on Services
Computing, pages 65–73. IEEE.
Total Cost Modeling for VNF based on Licenses and Resources
253