Price Modeling of IaaS Providers

An Approach Focused on Enterprise Application Integration

C´assio L. M. Belusso

, Sand ro Sawicki

, Vitor Basto-Fernandes

, Rafael Z. Frantz

and Fabricia Roos-Frantz

Federal University of Fronteira Sul, Cerro Largo, Brazil

Department of Exact Sciences and Engineering, UNIJU

I University, Iju´ı, Brazil

Department of Information Science and Technology, University Institute of Lisbon, Lisbon, Portugal

Keywords:

Cloud Computing, Enterprise Application Integration, Price Modeling, Linear Regression, IaaS.

Abstract:

One of the main advances in information technology today is cloud computing. It is a great alternative for users

to reduce costs related to the need to acquire and maintain computational infrastructure to develop, implement

and execute software applications. Cloud computing services are offered by providers and can be classiﬁed

into three main modalities: Platform-as-a-Service (PaaS), Software-as-a-Service (SaaS) and Infrastructure-

as-a-Service (IaaS). In IaaS, the user has a virtual machine at their disposal with the desired computational

resources at a given cost. Generally, the providers offer infrastructure services divided into instances, w ith pre-

established conﬁgurations. The main challenge faced by companies is to choose t he instance that best ﬁts their

needs among the many options offered by providers. Frequently, these companies need a large computational

infrastructure to manage and i mprove their business processes and, due to the high cost of maintaining local

infrastructure, they have begun to mi grate applications to the cloud in order to reduce these costs. In this

paper, we introduce a proposal for price modeling of instances of virtual machines using linear regression.

This approach analyzes a set of simpliﬁed hypotheses considering the following providers: Amazon EC2,

Google Compute Engine and Microsoft Windows Azure.

1 INTRODUCTION

Cloud computing is a signiﬁcant innovation in the

ﬁeld of informa tion technology as well as one of the

greatest perspectives for growth in the coming ye-

ars. Despite having gained much attention recently,

the concepts involved in this new computing resource

have been in development for a long time. However,

there is n ot yet a formal d eﬁnition for the concept of

cloud computing. Essentially, it is the utilization o f

the most diverse applications over the Internet, in the

same way as if they were installed on a computer or

other ph ysical device.

An imp ortant advantage of cloud computing is its

cost-beneﬁt relation, where the user pays only for

what is used (Murthy et al., 2012). In addition, users

of service s in the cloud are provided with mainte-

nance services, upgra des, backup, security and other

resources that would be needed were the applications

installed on their own comp uter.

Cloud computing services are offered by provi-

ders and there are several on the market offering va-

rious services, with emphasis on software services

(SaaS - Software-as-a-Service), where certain soft-

wares are provided and paid for on a per-use ba-

sis; platform services (PaaS - Platform-as-a-Service),

where the user is provided with an environment to de-

sign, test, and deploy custom applications; and infr a-

structure services (IaaS - Infrastructure-as-a-Service),

where the user is p rovided with virtual machines and

manages resources as desired (CPU, memory, storage,

data transfers, network band width, etc). Our research

focuses speciﬁcally on IaaS.

IaaS provider s offer a range of different plans

for certain instances, which vary according to the

needs of each client. Generally, individuals and even

small businesses require instances comprising small

demand s. Large companies, however, have large de-

mands, and thus h ave been mainly responsible fo r

more substantial investments in cloud computing.

One of the problems faced by users is that the cost

of cloud computing resources varies not only between

the numerous providers, but also between the diffe-

rent instances that each offers. This hinders decision

Belusso, C., Sawicki, S., Basto-Fernandes, V., Frantz, R. and Roos-Frantz, F.

Price Modeling of IaaS Providers - An Approach Focused on Enterprise Application Integration.

DOI: 10.5220/0006371603710376

In Proceedings of the 19th Inter national Conference on Enterprise Information Systems (ICEIS 2017) - Volume 1, pages 371-376

ISBN: 978-989-758-247-9

371

making, because often the lowest price is not always

the best choice, since the user may be looking fo r a

provider/instance that offers better Quality -of-Service

(QoS). For example, if users acquire an instance with

more resources th a n are needed, they will be sp e nding

money unnecessarily.

Instance prices charged by providers are based on

the amount of com putational resources employed by

users. However, a n important question attracting the

attention of research community is related to how

these prices are set. Whereas some resear ch shows

that there is a mechanism deﬁned for th is (Murthy

et al., 2012; Kihal et al., 2 012; Mitropoulou et al.,

2016), others show that price is much mo re complex

than was previously thought and that it is inﬂuenced

by economic policy and the well-known law of supply

and deman d (Al-Roomi et al., 2013; M azrekaj et al.,

2016). After prices are set by providers, the ﬁnal va-

lue of the instances can still change based on other

factors, such as the location wher e the virtual machine

is hosted or the operatin g system chosen by the user.

In the c ontext of Enterprise Application Integra-

tion (EAI), cloud computin g provides a high-ca pacity

computing infrastruc ture at a low cost, in which in-

tegration solutions can be deployed and run. Inte-

gration solutions are extremely important in the pro-

cess of integration, because they are softwares that

act as a communica tion link between th e different

applications contained in the software ecosystem of

compan ie s, allowing sharing inf ormation across them

quickly and efﬁciently (Frantz et al., 2016). Figure 1

illustrates the basic scheme of an integration solution.

Figure 1: Integration solution.

Despite large c ompanies having made signiﬁcant

investments in cloud computing to improve their bu-

siness processes, the providers do not offer a met-

hod to describe the variability of the services and

the constraints among them, so tha t these models can

be used in the decision-making proc ess (Hern´andez

et al., 2015). This study pr esents a new proposal for

modeling the instance prices charged u sin g linear re-

gression by three o f the main providers of clou d com-

puting: Amazon E C2 (Amazon, 2016), Google Com-

pute Engine (Google, 2016) and Microsoft Windows

Azure (Azure, 2016 ). Our aim is that this new model

can contribute to decision-making related to which

provider/instance best adapts to the needs of the com-

panies, resu lting in a signiﬁcant reduction of time and

money from the deployme nt and implementation of

integration solutions in the cloud. The proposal for

price modeling of instances of virtual m achines is

described by presenting some of the simp lifying as-

sumptions adopted in this new model.

The remainder of this article is organized as fol-

lows: Section 2 summarizes some papers that present

price modeling proposals for some providers on the

market, as well as some prospects for analysis of price

policy practiced by them; Section 3 presen ts a study

of the hypotheses fo r th e development of the new pro-

posal and theories chosen for the construction of the

model; Finally, in Section 4 the conclusions and some

prospects for future stu dies are presented .

2 RELATED WORK

The studies presented in this section seek to under-

stand how IaaS providers establish their pricing po-

licy for their services. It is widely understood that

this is not an easy task, which explains why there are

different approaches adopted.

Murthy et al. (2012) present a comparative study

of pricing and billing models of some IaaS providers.

The authors consider storage space as the main har-

dware re quirement, but CPU and memory are also

used to compare the instances. They also analyze dif-

ferences in the prices of instances for two different

operating systems and in the different forms of pur-

chasing o r contracting the service. On the other hand,

the hosting of virtual machines at different geograph i-

cal locations is not considered, even though this can

have a great inﬂue nce on prices.

Rodamilans (2014) states that the problem of se-

lection of instances can be solved b y characterizing

them and the applications, considering the ﬁnancial

cost and the performance of the virtu al machin e. He

proposes a method for selecting cloud provider s using

the following steps: characterization, selection an d

implementation. For this purpose, it is not conside-

red necessary to run the a pplication in all instances to

decide which is the most appropriate, rather it is only

necessary to identify the dominant computational re-

source.

Mazrekaj et al. (2016) highlight that the objective

of the providers in offering cloud computing servi-

ces is always to get the highest revenue through their

pricing schem es, while users seek the h ighest QoS

at a low pr ice. Thus, they make a price comparison

of some models and pricing schemes of cloud servi-

ces pr oviders based on the following aspects: serv ic e ,

quality, fair price and imp ortance in the market.

Chun and Choi (2014 ) analyze two pricing sche-

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

372

mes practiced by providers of cloud computing:

subscription and pay-per-use. However, unlike most

proposals studied, the author s make an analysis from

the per spective of the pr ovider. They claim that the

different pricing schemes are a consequenc e of the

providers’ different cost structures, which en tail ever-

ything from the investment in cloud infra structure to

the cost of a service being idle for a certain period of

time.

Kihal et al. (2012) present two methods to im-

prove the transparency of the prices charged by pro-

viders of IaaS. The ﬁrst is the hedonic pricing met-

hod, in which the fraction that each compo nent con-

tributes to the ﬁnal cost of the instance is calculated.

In this method, the instances n eed to be analyzed se-

parately, making it impossible to compare providers.

To work around this problem, the authors developed

the Pricing Plan Co mparison (PriCo), a method that

identiﬁes the best proﬁle for each provider. In both

cases, the authors considered as the main hypothe-

sis that customers seek the ch eapest service, not co n-

sidering, for example, the QoS. During the analy-

sis, the IaaS components memory, CPU and storage ,

Windows operating system and On-d emand instances

were used.

Mitropoulou et al. (2016) propose a price index

based on the hedonic pricing method, which accounts

for different factors in pricing models of cloud co m-

puting services at the IaaS level, including the geo-

graphic location of the provider. It is a mathema-

tical approach, speciﬁcally of regression models (li-

near, exponential, for co mparison), where the price of

a service is related to its characteristics. Using the

hedon ic method, the auth ors allowed for an adjust-

ment of the pr ic e of a service for the quality, not the

quantity. The authors collected data from providers

and analyzed in w hich regions they have hosted vir-

tual machines. The model also accounts for qualita-

tive variables related to the operatin g system and the

purcha sin g or contractual model. Quantitative vari-

ables include amo unt of memory, CPU, storage and

transfer

out. The authors did not consider some fac-

tors in the form of model variables because, according

to th e m, they did not have m uch inﬂue nce on the ﬁnal

price of the instances.

Our review of the latest research shows that th e re

are no proposals directly related to selection of IaaS

providers for migratio n of integration solutions to the

cloud. Likewise, instruments or methodologies ca-

pable o f quantifying the demand for computational

resources that an integration solution consumes were

not found. In this sense, our model can provide a nee-

ded new approach.

3 PROPOSED MODELING

In this paper, we present a proposal for modeling pri-

ces adopted by IaaS providers. The main focus is on

Enterprise Application Integration, speciﬁcally in in-

tegration solution migration to the cloud. D ue to this

peculiarity, some assumptions need to be established

and studied in this ﬁrst stage, considering the aims

and target audience.

The objective assumed in modeling is to always

obtain the instance with the lowest price, but with a

QoS c apable of per forming the prede te rmined com-

putational demand. In addition, some factors inﬂue n-

cing prices are analyzed, as well as the ne ed to build

them into the model.

3.1 Simplifying Hypotheses

Among the many IaaS providers on the market, in this

study we opted to analyze three of them: Amazon

EC2, Google Compute Engine and Microsoft Win-

dows Azure. This choice was based on their popu -

larity and , in the case of Amazon EC2, also due to the

complexity of instance pricing. Some proposals are

able to analyze more than one provider at the same

time, and this is considered to be a Cross-Provider

compariso n (Hern´andez et al., 2015).

For this pur pose, simplifying hypotheses are nee-

ded, initially, for some factors that inﬂu ence, directly

or indirectly, the prices c harged b y providers. These

include:

• The Operating System: opting for a virtual ma-

chine that uses the Linux op erating system as op-

posed to Windows, fo r example, can result in a

signiﬁcant difference in the p rice of the instances.

For so me providers, the price may also vary if the

platforms are different. Information collected di-

rectly fro m the Internet homepages of providers

indicates that an instance is always cheaper when

opting for Linux.

• Instance Type: depending on the requirements

for the deploym ent and implementation of inte-

gration solutions, instances with smaller or larger

conﬁgurations, or instances of a High -Memory

or High-CPU type, may be c hosen, all of them

grouped according to particular hardware requi-

rements. High-Memory instances have large

amounts of memory, while the amount of other

components of the instance is small o r has small

variations. Similarly, High-CPU instances h ave

a large processing capacity. Applicatio n integra-

tion solutions are light a nd don’t require a lot o f

storage space. Thus, this requirement is not con-

sidered. Some providers have instance s that differ

Price Modeling of IaaS Providers - An Approach Focused on Enterprise Application Integration

373

with regard to the storage unit, wh ic h ca n be Hard

Disk Drive (HDD) or Solid-State Drive ( SSD) ,

and this can lead to changes in price. There is

also grouping of instances considering th e users’

proﬁles. For those who use little, providers o ffer

instances with a smaller conﬁgurations, while the

instances with more powerful conﬁgurations are

used by large companies, which have a large vo-

lume of applications running simultaneously.

• Discount Mo del: generally, the providers offer

this to users who acquire an instance for a g iven

amount of time. The longer the time of acqui-

red, the greater the discount will b e for the usage

rate. There a re also cases whe re the user can pay a

percentage of the total cost in advance, where the

higher the perc entage of this payment, the g rea-

ter the discount obtained. Other fo rms of discount

include that for allocation of virtual machines in

speciﬁc regions (Amazon, 2016), for pe rcentage

of month ly use of the virtual machine (Google,

2016) and for instances with cost above a pre-set

value (Azure, 2016). Due to the d ifferent types of

discounts offered, the elaboration of a model that

includes all of them is a complex task , considering

the peculiarities of each. For example, consider a

provider P, for which one o f the discount modes is

given for the amount paid, and two clients, A and

B, which have spe nt the same amount of money

and, so, theore tica lly would get the sam e discoun t.

Assuming that client A has been u sing the service

provider P for a lo ng time while client B is using

the service for the ﬁrst time. Customer A may

be offered a discount greater due to their allegi-

ance, thus partaking of another form of discount.

Therefore, considering this imminent difﬁculty in

modeling discounts, they are not considered in the

proposed mod el.

• Geographical Location of the Provider: many

providers have virtual machines that are hosted in

various parts of the world and the price of the in-

stance can change considerably within the same

provider. Some providers have more than 10 pos-

sible location s (Amazon , 2016; Azure, 2016), ma-

king it difﬁcult to compare. However, users may

not always be a ble to choose the location in which

the instance is cheaper, because there may be legal

issues of the co untry in which the virtual machine

is hosted. Mitr opoulou et al. (2016) analyzed the

inﬂuence of geographical location on the prices

charged by providers a nd choose to group the dif-

ferent regions covered by continent. I n the present

proposal, which foc uses on the lowest price with

an ideal QoS, an adja c ent survey of each of th e

providers should be made in order to iden tify re-

gions with the lowest prices, limited to, at most,

ﬁve of them.

• Billing Model: in addition to having numerous in-

stances composed of different conﬁgurations, the

providers also offer different pricing models. For

example, Amazon EC2 provides three pricing mo-

dels: On-demand, Reserved and Spot. For On-

demand instances, the user pays according to use,

without a long-term commitment; in Reserved in-

stances, the user acquires the instance fo r a given

amount of time and, because of this, pays a lo-

wer usage rate (per hour or per minute); ﬁnally,

Spot instances allow the user to take par t of a kind

of auction for unused c omputing capacity, which

can generate savings of up to 90 % in relation to

the On-dema nd mode. However, the spot price

ﬂuctuates based on supply and demand for avai-

lable capacity. If the user’s offer is over that of

the spot price, the instance will c ontinue running;

otherwise, the serv ic e will be inte rrupted. The

need to stand out on the cloud computing mar-

ket makes providers offer different pricing mo-

dels, among which the On-deman d model and the

Reserved mod e l are the most common. The Re-

served models are always more economical in the

long term and the price tends to fall even more as

the subscrip tion time increases (an analysis of pri-

cing models Amazon EC2 can be found at Murthy

et al. (2012)).

• Period of Use: ﬁnally, a more detailed study seeks

to identify whether all of the chosen p roviders

change their prices for the use of the serv ic e s at

different times and whether this alteration is sig-

niﬁcant. T he user may choose to run applications

on different schedules to achieve a reduction in the

cost of their virtual machine. At peak hours, the

price of the instances tends to be larger or the dis-

count offered, if any, ma y be lower. Each provider

deﬁnes peak h ours.

3.2 Mathematical Model

Multiple linear regression was used to estimate the

price of the instances for the three selected providers,

due to the linear behavior of the data collected fr om

the providers. Durin g data collection, a ll the possibi-

lities of geographical location of each of the providers

were co nsidered, as w ell as information pertain ing to

Linux a nd Win dows operating systems. The pricing

model considere d was the On-demand model.

Using the regression method, it is possible to es-

timate the c ost of individual c haracteristics that in-

ﬂuence the ﬁnal price of the instances by ca lc ulating

the coefﬁcients for each of the variables of the model.

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

374

The multiple linear regression m odel adapted to this

proposal is described by Equation 1.

= a

+ a

+ ... + a

+ ε

, i = 1, ..., n

(1)

where,

• i refers to the i-th instan c e of any one of the three

chosen providers;

• n is the number o f variables of the model;

• C

is the ﬁnal price of the instance i;

• a

are the co e fﬁcients of regression to be calcula-

ted, i.e. the sh are of participation of feature X in

the ﬁnal price of the instance i;

• X

are the independent variables of the model,

that is each of the characteristics that inﬂuence the

ﬁnal price of the instance i;

• ε

is the residual error of the regression for each

instance i.

The model variables can be summarized as quanti-

tative variables, composed by hardware requirements

(for example, CPU, memory and storage), and quali-

tative variables, consisting of software requirements

(for example, operating system and platform) and ot-

her variables (for example, geographical location).

Qualitative variables of the model are represented

using Dummy variables, where a certain trait can take

on the value 0 or 1 (Wonnacott and Wonnacott, 1990).

The applicatio n of the model can be better un der-

stood from the following example:

Example: an integration solution requires 4 GB of

RAM, 2 processing cores and 200 MB of storage to

be executed. By inserting this demand into th e model,

the price of all instanc es able to perform this solu tion

is calculated and the cheapest instance is ind ic ated.

The Dummy variables are deﬁned and inserted previ-

ously in to the model. For the operating system, for

example, the value of 1 f or Windows and 0 for Li-

nux can be set. If the option is Windows, multiplying

the coefﬁcient from the regression m odel for 1; being

Linux, the coefﬁcient is multiplied by 0. Assuming

that an instance is more expensive when you choose

Windows, the coefﬁcient of this variable should be

positive, generating a n increase in the ﬁnal price of

the instance ; in the case of the instance being chea-

per, the coefﬁcient should be negative, reducing the

price. The same process can be adopted for th e other

qualitative variables. The coe fﬁcients of the quanti-

tative variables represent the unit price of each of the

hardware components, i.e. the c ost per unit of RAM,

CPU and storage.

Because it is a method o f estimation, the level of

reliability of the linear model must be considered. I n

addition, most models feature non-linearity and, be-

cause of that, a no n-linear model may be more ade-

quate. However, in these cases, due to difﬁculties

encountered in non-linear modeling, the alternative

used by researchers is the linearization of the varia-

bles, which sometimes does not cause major distorti-

ons to the results.

4 CONCLUSIONS

This study proposed a scheme for modeling the in-

stance prices charged by providers of cloud compu-

ting at the IaaS level, using a linear regression model

and some simplify ing hypotheses. This proposal con-

sidered the providers Amazon EC2, Google Compute

Engine and Microsoft Windows Azure.

The literature review show that no proposal to date

has been geared toward implementatio n and execu-

tion of integration solutions in the cloud. The ab-

sence of a mechanism that can offer companies the

opportunity to compare the servic es offered by provi-

ders makes this a new and prom isin g approach, con-

sidering that, curr ently, all information must be col-

lected directly from the Internet homepage of provi-

ders, delayin g the decision-making process and, con-

sequently, making this task quite onerous.

For future studies, it is necessary to ﬁrst obtain th e

coefﬁcients of the regression model and verify their

level of reliability. Complem e ntarily, it is possible

to build a m ethod of estimatio n able to quantify the

demand for computational infrastructure necessary to

perform an integration solution.

ACKNOWLEDGEMENTS

The research work on which we report in this paper

is supported by CNPq, CAPES, FAPERGS, and the

internal Research Programme at UNIJUI University.

First author is also thanks the Federal University of

Fronteira Sul (UFFS) for the support to the develop-

ment of his research.

REFERENCES

Al-Roomi, M., Al-Ebrahim, S., Buqrais, S., and Ahmad, I.

(2013). Cloud computing pricing models: a survey.

International Journal of Grid and Distributed Com-

puting, 6(5):93–106.

Amazon (2016). Amazon web services.

https://aws.amazon.com/. Acesso em: 28/12/2016.

Price Modeling of IaaS Providers - An Approach Focused on Enterprise Application Integration

375

Azure (2016). Microsoft windows azure.

https://azure.com/. Acesso em: 24/12/2016.

Chun, S. and Choi, B . (2014). Service models and pricing

schemes for cloud computing. Cluster Computing,

17(2):529–535.

Frantz, R. Z., Corchuelo, R., and Roos-Frantz, F. (2016).

On the design of a maintainable software development

kit to implement integration solutions. The Journal of

Systems and Software, 111:89–104.

Google (2016). Google cloud plataform.

https://cloud.google.com/. Acesso em: 25/12/2016.

Hern´andez, I., Sawicki, S., Roos-Frantz, F. , and Frantz,

R. Z. (2015). Cloud conﬁguration modelling: a li-

terature review from an application integration de-

ployment perspective. Procedia Computer Science,

64:977–983.

Kihal, S. E., Schlereth, C., and Skiera, B. (2012). Price

comparasion for Infrastructure-as-a-Service. In Eu-

ropean Conference on Information Systems. Resear-

chGate.

Mazrekaj, A., Shabani, I., and Sejdiu, B. (2016). Pricing

schemes in cloud computing: an overview. I nternati-

onal Journal of Advanced Computer Science and Ap-

plication, 7(2):80–86.

Mitropoulou, P., Filiopoulou, E., Michalakelis, C., and Ni-

kolaidou, M. (2016). Pricing cloud IaaS services ba-

sed on a hedonic price index. Computing, pages 1–15.

Murthy, M. K. M., Sanjay, H. A., and Padmanabha, A. J.

(2012). Pricing models and pricing schemes of IaaS

providers: a comparison study. In Proceedings of

the International Conference on Advances in Com-

puting, Communications and Informatics, pages 143–

147. ACM.

Rodamilans, C. B. (2014). Uma metodologia para

caracterizac¸ ˜ao de aplicac¸ ˜oes e de instˆancias de

m´aquinas virtuais no ambiente de computac¸ ˜ao em nu-

vem. Tese de doutorado, Escola Polit´ecnica da Uni-

versidade de S˜ao Paulo, S˜ao Paulo.

Wonnacott, T. H. and Wonnacott, R. J. (1990). Introductory

Statistics for Business and Economics. John Wiley &

Sons, New York, 4th edition.

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

376