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

net/publication/262326134

Cost models-pillars for efficient cloud computing: Position paper

Article  in  International Journal of Intelligent Systems Technologies and Applications · July 2013

DOI: 10.1504/IJISTA.2013.055102

CITATIONS READS

6 1,335

2 authors:

Catalin Negru Valentin Cristea

Polytechnic University of Bucharest Polytechnic University of Bucharest
41 PUBLICATIONS   152 CITATIONS    281 PUBLICATIONS   1,981 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Data4Water View project

DataWay - Real-time Data Processing Platform for Smart Cities: Making sense of Big Data View project

All content following this page was uploaded by Catalin Negru on 17 September 2014.

The user has requested enhancement of the downloaded file.

28 Int. J. Intelligent Systems Technologies and Applications, Vol. 12, No. 1, 2013

Cost models – pillars for efficient cloud computing:

position paper

Catalin Negru* and Valentin Cristea

Faculty of Automatic Control and Computer Science,
University POLITHEHNICA of Bucharest,
Splaiul Independentei nr. 313,
sector 6, Bucuresti RO-060042, Romania
E-mail: [email protected]
E-mail: [email protected]
*Corresponding author

Abstract: Cost models are fundamental building blocks for cloud computing.
Cloud providers offer a very wide portfolio of services, while Cloud clients
access them against some financial arrangement. There is a fundamental trade-
off between what Cloud provider can offer in terms of resources (software or
hardware), services (e.g., storage, e-mail) cost and what Cloud clients are
willing to pay. The Cloud cost models today are distinct based on service
models: Infrastructure as a Service (IaaS), Platform as a Service (PaaS) and
Software as a Service (SaaS). The need to forecast the cost over a period
of time imposes building of cost models, which have to be accurate and
error free. In this paper, we make a survey and analyse existing cost models in
cloud computing and discuss open issues related to the subject.

Keywords: cost models; cloud computing; efficiency; optimisation.

Reference to this paper should be made as follows: Negru, C. and Cristea, V.

(2013) ‘Cost models – pillars for efficient cloud computing: position paper’,
Int. J. Intelligent Systems Technologies and Applications, Vol. 12, No. 1,
pp.28–38.

Biographical notes: Catalin Negru is currently a System Engineer and

PhD student within the Computer Science Department, in Politehnica
University of Bucharest, Romania. His general interests regard cloud
computing. His current research is focused on optimising the cost for cloud
storage systems.

Valentin Cristea is a Professor of the Computer Science and Engineering

Department of University POLITEHNICA of Bucharest. His main fields of
expertise are distributed systems, grid computing and e-services. In 2003 and
2011 he received the IBM faculty award. He published more than 120 specialist
papers, 25 books (in Romanian, English, and French), and 60 technical reports.
He is PhD supervisor, coordinator of the Master program on Advanced
Software Services, and the Romanian coordinator of the Master program on
Parallel and Distributed Computer Systems co-developed with VU Amsterdam.

Copyright © 2013 Inderscience Enterprises Ltd.

 Cost models – pillars for efficient cloud computing: position paper 29

1 Introduction

Cloud computing paradigm permits to the users to use computational and software
resources in a pay-per-use manner. In this way the user has a great flexibility to adapt to
changes. In cloud computing everything is provided as a service to cloud user, in a
functional, usable and extremely powerful manner, meaning that cloud clients can access
the services almost everywhere if they have internet connection and have access to a
large number of resources. Companies like Google, Facebook, Yahoo! and Microsoft
have invested in their data centres to support cloud services, the natural question that
rises for that fact is the following one: where does the cost go in today’s cloud service
data centres?
There are three principal types of services that can be offered in cloud computing:
Infrastructure as a Service (IaaS), Platform as a Service (PaaS) and Software as a Service
(SaaS). IaaS assume provisioning with fundamental computing resources such as
processing, storage, networks and so on, to the consumer which can deploy and run
arbitrary software, meaning middleware software, operating systems and others. The
consumer has control over operating systems, storage and deployed applications but does
not manage or control the underlying cloud infrastructure (Armbrust et al., 2010).
PaaS offer to the consumer the capability to deploy onto the cloud infrastructure
applications created using programming languages, tools and libraries supported by the
provider. The consumer has control over the deployed applications and possibly
configuration settings for the application-hosting environment (Armbrust et al., 2010).
SaaS is the capability provided to the consumer to use the provider’s applications
running on a cloud infrastructure. The applications are accessible from various client
devices through either a thin client interface, such as a web browser (e.g., web-based
e-mail), or a programme interface (Armbrust et al., 2010).
Cost structure represents the identification of how costs associated with the
production of a good or service is distributed through the process (Martens et al., 2012).
Is important to identify the cost factors that determine the overall cost for offering a
certain service in cloud computing (Kondo et al, 2009).
Another important aspect that must be taken into consideration is represented by the
metrics that cloud providers should use to quantify their investment, and to calculate the
profitability for a certain service that is offered.
The remainder of this paper is organised as follows: Section 2 present the most
important cost factors in cloud computing. In Section 3, we present metrics used in
cost models to calculate the profitability of investment in a cloud computing platform.
Section 4 review existing cost models in cloud computing. Section 5 presents different
pricing strategies used in cloud computing. Section 6 presents different cost reduction
strategies used today in cloud platforms to optimise the utilisation of a cloud platform.
Finally, Section 7 presents a discussion related to cost models and draws a few
conclusions.

2 Important cost factors in cloud computing

In this section, we present the most important factors that are included in the structure of
the cost for a cloud platform. There are two sides of the same problem: the cloud provider
perspective and cloud user perspective. The cloud provider must calculate his total cost of
30 C. Negru and V. Cristea

ownership and adequately put price on his services to have profit and amortise his
investment and on the other side cloud user has to calculate total cost of running his
application in the cloud so there are cost factors that regard one side or another.
Table 1 gives the costs for the service user, and also for the service provider.

Table 1 TOC perspective

Component IaaS PaaS SaaS

Business process Customer Customer Customer
Business logic Customer Customer Service provider
Middleware management Customer Service provider Service provider
Application licensing/Support Customer Service provider Service provider
OS management Customer Service provider Service provider
OS licensing/Support Service provider Service provider Service provider
Server/Storage/Networks Service provider Service provider Service provider
HW/Maintenance
Utilities Service provider Service provider Service provider
ME equipment Service provider Service provider Service provider
Real estate Service provider Service provider Service provider
Source: Rutland (2012)

The authors of Kashef and Altmann (2011) identify five groups of cost factors that
mainly regards cloud provider and partially to cloud users: electricity, hardware,
software, labour and business premises.
The electricity cost factor refers to: electricity for cooling infrastructure, for powering
computing and networking devices and other electronic devices. Also must be taken into
consideration the two values of power consumption: the value when the system is idle
and when the system is heavily used.
The hardware cost factor refers to the acquisition of hardware and networking
resources, needed in-house. Because the all the resources have an optimal lifetime for
exploitation, a depreciation parameter must be taken into consideration to evaluate the
amortisation of the investment. Software cost refers to the price that cloud provider have
to pay to purchase the software licenses for server operating systems, middleware
software, application software and are applicable to a SaaS provider.
Labour cost factor include salaries for technicians who work for maintenance of the
data centre and in the support area. This cost varies in function of the location of the data
centre.
Business premises cost factor includes collateral cost such as: the price for data centre
facility, price for all non-electric instruments, the price for cabling and so on.
A consumer of cloud services must take into consideration the following factors to
calculate his total cost for running an application: the usage cost for servers (CPU hours),
the cost for incoming data transfer in the cloud platform, cost for the amount of outgoing
data transfer which are measured in Giga Bytes per seconds (GB/s), cloud storage factor
measured in Giga Bytes (GB), the cost for executing input requests in cloud platform, the
cost for executing an output requests from cloud platform.
 Cost models – pillars for efficient cloud computing: position paper 31

3 Important metrics in cost models

To calculate the profitability of investment, the cloud provider must perform financial
analysis with the aid of a few parameters that can be calculated along with Return of
Investment (ROI) such as Net Present Value and Internal Rate of Return (Tak et al.,
2011).
Net Present Value (NPV) is defined as the difference between the present value of
cash inflows and the present value of cash outflows. NPV is used in capital budgeting to
analyse the profitability of an investment or project. The formula for NPV
(http://en.wikipedia.org/wiki/Net_present_value) is presented below:
N
Rt
NPV(i, N ) = ∑ , (1)
(1 + i )
t
t =1

where t is the time of the cash flow; i is discount rate (the rate of return that could be
earned on an investment in the financial markets with similar risk); the opportunity cost
of capital; Rt is the net cash flow (the amount of cash, inflow minus outflow) at time t.
The Internal Rate of Return (IRR) is a parameter for measuring the financial
evaluation. It is the discount rate for which a project’s benefits exactly equal its
costs, meaning that, it is the rate at which the project’s net present value is zero.
The evolution of IRR parameter is shown in Figure 1. As can be seen a higher IRR
parameter make an investment more desirable, and exceeding the IRR value, will produce
business loss.

Figure 1 Usual internal rate of return (see online version for colours)

Equation (2) describes the IRR parameter:

T
Bi
∑ − C0 = 0. (2)
(1 + i )
t
t =1

ROI represent an important indicator that helps in decision of moving or not in to the
cloud, by measuring the efficiency of an investment. When calculating ROI a large
number of factors involved in a business are taking into consideration. For calculating
ROI, the following information’s are needed: the initial cost of project, the investment
made, the cost savings done owing to the new investment (Misra and Mondal, 2011).
32 C. Negru and V. Cristea

The ROI formula for cloud computing with time frame per month or per year is the
following (Misra and Mondal, 2011):
( Initial cost − Final cost ) − Investment Costs saved − Investment
ROI = = .
Investment Investment

4 Different cost models in cloud computing

The authors of Truong and Dustdar (2010) present basic cost models which are shown
in Table 2. The first four models, Mds (cost model for data storage); Mcm (cost model
for computational machine); Mdfi (cost model for data transfer into the cloud) and Mdfo
(cost model for data transfer out to the cloud) are provided by cloud providers in their
pricing specifications. Utilising presented cost models and monitoring data the authors
develop a cost estimation, monitoring and analysis service for the scientific application
that run in cloud environments. The architecture of the solution is present in Figure 2.
The shortcoming of this solution is represented by the fact that is applicable only to
scientific applications and the model does not take into consideration the workload of the
application (how many users utilise the service or application at one time) and therefore
cannot be applicable to other types of applications or services for example a mail service
or a website.
The authors of Kashef and Altmann (2011) proposed a cost model for hybrid cloud
(i.e., the combinations of a private data centre (private Cloud) and the public Cloud).
They present a conceptual model that assumes an organisation which comprises the
execution of N applications and M services. Cloud users buy services to construct their
applications. For the cost of data communication is used a directed weighted graph which
is shown in Figure 3, where edges show data communication and vertices represent
services. On the basis of the graph is constructed a distance matrix, which represent the
data-transfer related cost factors ai,j between service i and j. The authors propose a cost
formula which is the sum of two types of cost: a fixed cost based on cost factors
presented earlier and a variable cost based on the services used for cloud provider. The
problem with this approach is that it takes into consideration direct usage of resource and
do not take into account the Burstiness of the services, and unallocated resources. This
problem in solved by the authors of Gmach et al. (2011) by taking into account the
Burstiness of the applications and unallocated resources.
Another important aspect for cost models in cloud computing is represented by
virtualisation technology, which impose supplementary challenges for IaaS providers to
estimate cost and for SaaS providers to bill.
In Gmach et al. (2011), present three models for apportioning cost in a virtualised
environment and indicate the one that provide the most robust and repeatable cost
estimates. First, they consider a server-usage model that takes into account only the direct
resource consumption by W workloads, with cost Cs of server s, cost that is composed by
CAPEX component (e.g., fraction of acquisition costs based on the length of the
considered interval) and OPEX component (e.g., costs for power associated with the
server). They define the following server workload:
 Cost models – pillars for efficient cloud computing: position paper 33

server-usage
ds,w
∏ = Cs . (3)
s,w ∑ w′=1
d s , w′

To take into account Burstiness and unallocated resources the authors partition server
cost Cs based on utilisation to get Cds, Cbs, Cas, respectively, where Cds corresponds to
costs associated with the average utilisation of the server s, and Cbs and Cas correspond to
the difference between peak and average utilisation of the resource, and difference
between 100% and the peak utilisation of the resource, respectively.
Another model is server-burst model that divide the burst portion of cost for a server
in a manner that is weighted by the Burstiness of each workload on the server.
In a second step, unallocated resources of the server are apportioned based on the bursty
costs (Gmach et al., 2011). The server burst is defined as:
server-burst-temp
ds,w ε + bs , w
∏ = Csd + Csb (4)
∑ ∑
W W
s,w
w′ =1
d s , w′ w′ =1
(ε + bs , w′ )

∏
server-burst-temp
server-burst server-burst-temp

∏ = ∏ +C a s,w
. (5)
∑ ∏
s W server-burst-temp
s,w s,w
w′=1 s,w

Table 2 Costs te and te(p) are the total elapsed time for executing computational task or data
transfer and the execution time obtained with p parallel processes/threads,
respectively. n is the number of machines used or to be used

Model Activities Cost

Mds Data storage size(total) × tsub × cost (storage), where tsub is
subscription time
Mcm Computational machine cost(machine)
Mdfi Data transfer into the cloud cost(transferin)
Mdfo Data transfer out to the cloud cost(transferout)
Mds Single data transfer without the cost size(in) × Mdfi + size(out) × Mdfo
for machines performing the transfer
Msm Sequential/multi-threaded program tc × Mcm + size(out)XMdfo + size(in) × Mdfi
or single data transfer with the cost
for machines performing the transfer
(cost monitoring)
Mse Sequential or multi-threaded fpi × Mcm + size(out) × Mdfo + size(in) × Mdfi, where
program (cost estimation) fpi is an estimated performance improvement
function can be provided by performance
prediction tools or scientists. In our case, currently,
an ideall parallel performance improvement
fpi = p/n × tc(p) where p is the number of threads
used to obtain tc(p), p and tc are known knowledge
Mpm Parallel/MPI programs on multiple n × Mcm × tc + size(out) × Mdfo + size(in) × Mdfi
machines (cost monitoring)
Mpe Parallel/MPI programs on multiple n × Mcm × fpi + size(out) × Mdfo + size(in) × Mdfi
machines (cost estimation) where fpi is an estimated improvement function
when n processes are used
34 C. Negru and V. Cristea

Figure 2 Cloud cost estimation and monitoring service

Source: Truong and Dustdar (2010)

Figure 3 Direct weighted graph

Source: Kashef and Altmann (2011)

The third model proposed is pool-burst model Burstiness cost and unallocated
resources using measures for the S servers in the resource pool instead of the individual
servers:
pool-burst-temp
ds,w ⎛ S ⎞ ε + bs , w
∏ = Csd + ⎜ ∑ Csb ⎟ S ,W (6)
∑ ⎝ s ′=1 ⎠ ∑ s ′ =1, w′ =1 ε + bs ′, w′
W
s,w
w′ =1
d s , w′

∏
pool-burst-temp
pool-burst pool-burs-temp
⎛ S ⎞
∏ = ∏ + ⎜ ∑ Csa′ ⎟ S ,W s , w pool-burst-temp . (7)
s.w s,w ⎝ s′=1 ⎠ ∑ s′ =1, w′ =1 ∏ s′, w′

The cost models presented in Gmach et al. (2011) take into consideration only the server
usage cost. The cost for data communication of the application is not taken into
consideration so therefore is not a complete cost model.
 Cost models – pillars for efficient cloud computing: position paper 35

From the cost models presented above namely that one of them are tailored only to
HPC applications, others take into consideration only the server cost and ignore the
communication costs, raises clearly the necessity for a general cost model that can
accurate predict the cost for running an application in cloud environment.
In Table 3, present the drawbacks of existing cost models.

Table 3 Drawback of existing models

Cost model Advantage Disadvantages

Model for data storage These cost models can be Those models does not take
Model for computational machine combined, in order to into consideration the workload
obtain a general cost of the application (how many
Model for data transfer into the model users utilise the service or
cloud application at one time)
Model for data transfer out to the
cloud
Cost model for hybrid cloud Designed for applications Do not take into account the
that run in a public and Burstiness of the services, and
private cloud unallocated resources
Server-usage model Take into consideration Cost for data communication is
Server-burst model the usage pattern of the not considered
server
Pool-burst model

5 Pricing strategies in cloud computing

Another important aspect in cost models is represented by pricing mechanisms.

There are three categories of pricing mechanisms: fixed pricing, differential pricing
and market pricing. In the fixed pricing category are the following billing options:
pay-per-use, subscription and list price. For differential pricing are the following: service
feature dependant, customer characteristic dependant, volume dependant and value based.
In the market pricing category are three types of pricing mechanisms: bargaining, auction
and dynamic market (https://developers.google.com/appengine/docs/billing).
For example Google use a combined approach, for different services, where the client
can choose a pay-per-use for IaaS service model (the user pay CPU/hour), or in the case
of SaaS Google offer a subscription model. The volume dependant pricing mechanism is
used especially in cloud storage, where the client can have a better price if store a larger
amount of storage (Niu et al., 2012).
The authors of Macías and Guitart (2011) propose a genetic algorithm. They start
with a naive pricing function that evolves to a pricing function that offers suitable prices
in function of the system status. The proposed model is evaluated by comparison with the
following pricing strategies: fixed pricing, random pricing, utility maximisation price.
The results obtained are shown in Figure 4.
As can be seen the cloud provider that use random-pricing (prices are offered
randomly, in a uniform distribution, between buyer and seller) strategy is most
inefficient. In case of fixed pricing (offered prices are the 5% between the minimum price
that the provider can offer to not lose money and the maximum price that the client can
pay to get benefit by buying the service) the revenue of the provider increase linearly
36 C. Negru and V. Cristea

with the number of tasks. In conclusion Figure 2 shows that the provider using genetic
algorithm gets the highest revenue in most of the scenarios. When the maximum number
of tasks is high, both solutions are similar.

Figure 4 Comparison of revenues between four types of pricing. A provider with a flexible
genome (200 chromosomes and 6% of mutations) is used (see online version
for colours)

Source: Macías and Guitart (2011)

6 Discussion and conclusions

From the in formations presented above we conclude that a major problem with cost
models today are that there is not a general optimised model for estimating the costs, and
is difficult to estimate cost factors, and their variation over time.
At data centre level cost are concentrated in servers, power infrastructure, networking
and requirements. A low utilisation of this resources leads to very low efficiency and
business loss. Also the power consumption has a very important role in the efficiency of
the data centre and the reduction of costs. Geo-diversifying the location of data centres
(Peterson et al., 2011) can improve performance and increase reliability in the event of
site failures, and also reduce the costs, by buying cheaper labour for example (Greenberg
et al., 2008).
A cost model that take into consideration the characteristics of applications or service
such as: the data pattern, the data transfer, the average and peak utilisation (Yao et al.,
2012), would be more realistic one and help to estimate more accurate the cost for
running in the cloud. Also a good estimation of variable cost factors is needed; contrary
the cost model will give poor results and estimations.
An important aspect is represented by the major cost factors and their proper
estimation, since any error will have a major impact on the accuracy of the overall
estimated cost. To be more precise, accurately focusing on major cost factors is
suggested, since any error in their estimates has a large impact on the accuracy of the
 Cost models – pillars for efficient cloud computing: position paper 37

overall cost estimation. For instance, in many cases, 31% of cost of data centres comes
from labour, 30% from servers, and 25% from cooling (Greenberg et al., 2008).
A low-quality estimate on those cost factors has a large impact on the overall cost.

References
Armbrust, M., Fox, A., Griffith, R., Joseph, A.D., Katz, R., Konwinski, A., Lee, G., Patterson, D.,
Rabkin, A., Stoica, I. and Zaharia, M. (2010) ‘A view of cloud computing’, Commun. ACM,
Vol. 53, No. 4, April, pp.50–58.
Gmach, D., Rolia, J. and Cherkasova, L. (2011) ‘Resource and virtualization costs up in the cloud:
Models and design choices’, Proceedings of the 2011 IEEE/IFIP 41st International
Conference on Dependable Systems & Networks (DSN ‘11), IEEE Computer Society,
Washington DC, USA, pp.395–402.
Greenberg, A., Hamilton, J., Maltz, D.A. and Patel, P. (2008) ‘The cost of a cloud: research
problems in data center networks’, ACM SIGCOMM Computer Communication Review,
Vol. 39, No. 1, pp.68–73.
Kashef, M.M. and Altmann, J. (2011) ‘A cost model for hybrid clouds’, in Vanmechelen, K.,
Altmann, J. and Rana, O.F. (Eds.): Proceedings of the 8th International Conference on
Economics of Grids, Clouds, Systems, and Services (GECON’11), Springer-Verlag, Berlin,
Heidelberg, pp.46–60.
Kondo, D., Javadi, B., Malecot, P., Cappello, F. and Anderson, D.P. (2009) ‘Cost-benefit analysis
of Cloud Computing versus desktop grids’, Proceedings of the 2009 IEEE International
Symposium on Parallel & Distributed Processing (IPDPS ‘09), IEEE Computer Society,
Washington DC, USA, pp.1–12.
Macías, M. and Guitart, J. (2011) ‘A genetic model for pricing in cloud computing markets’,
Proceedings of the 2011 ACM Symposium on Applied Computing (SAC ‘11), ACM,
New York, NY, USA, pp.113–118.
Martens, B., Walterbusch, M. and Teuteberg, F. (2012) ‘Costing of cloud computing services:
a total cost of ownership approach’, Proceedings of the 2012 45th Hawaii International
Conference on System Sciences (HICSS ‘12), IEEE Computer Society, Washington DC, USA,
pp.1563–1572.
Misra, S.C. and Mondal, A. (2011) ‘Identification of a company’s suitability for the adoption of
cloud computing and modelling its corresponding Return on Investment’, Math. Comput.
Model., Vol. 53, Nos. 3–4, February, pp.504–521.
Niu, D., Feng, C. and Li, B. (2012) ‘Pricing cloud bandwidth reservations under demand
uncertainty’, Proceedings of the 12th ACM SIGMETRICS/PERFORMANCE Joint
International Conference on Measurement and Modeling of Computer Systems (SIGMETRICS
‘12), ACM, New York, NY, USA, pp.151–162.
Peterson, Z.N.J., Gondree, M. and Beverly, R. (2011) ‘A position paper on data sovereignty:
the importance of geolocating data in the cloud’, Proceedings of the 3rd USENIX conference
on Hot topics in cloud computing (HotCloud’11), USENIX Association, Berkeley, CA, USA,
p.9-9.
Rutland, D. (2012) In Cloud: Economics, Retrieved 15 October, 2012 from http://c1776742.cdn.
cloudfiles.rackspacecloud.com/downloads/pdfs/WhitePaper_CloudEconomics.pdf
Tak, B.C., Urgaonkar, B. and Sivasubramaniam, A. (2011) ‘To move or not to move: the
economics of cloud computing’, Proceedings of the 3rd USENIX Conference on Hot Topics in
Cloud Computing (HotCloud’11), USENIX Association, Berkeley, CA, USA, p.5-5.
Truong, H-L. and Dustdar, S. (2010) ‘Composable cost estimation and monitoring for
computational applications in cloud computing environments’, Procedia Computer Science,
Vol. 1, No. 1, May, pp.2175–2184, ISSN 1877-0509, 10.1016/j.procs.2010.04.243.
 38 C. Negru and V. Cristea

Yao, J., Liu, X., He, W. and Rahman, A. (2012) ‘Dynamic control of electricity cost with power
demand smoothing and peak shaving for distributed internet data centers’, Proceedings of the
2012 IEEE 32nd International Conference on Distributed Computing Systems (ICDCS ‘12),
IEEE Computer Society, Washington DC, USA, pp.416–424.

Websites
http://en.wikipedia.org/wiki/Net_present_value
https://developers.google.com/appengine/docs/billing

View publication stats