LATEX.EXAMPLE

 % This is LLNCS.DEM the demonstration file of
% the LaTeX macro package from Springer-Verlag
% for Lecture Notes in Computer Science,
% version 2.4 for LaTeX2e as of 16. April 2010
%
\documentclass{llncs}
%
\usepackage{makeidx}  % allows for indexgeneration
\usepackage{llncsdoc}
\usepackage{remreset}
\usepackage{aliascnt}
\usepackage{etoolbox}




\usepackage{fancyhdr} % to change header and footers
\usepackage{blindtext} 
\usepackage{tabularx} 
\usepackage{epigraph}
\usepackage{csquotes}
\usepackage{tikz}
\usepackage{pgf-pie}
\usepackage{makeidx}  % allows for indexgeneration
\usepackage{llncsdoc}
\usepackage{remreset}
\usepackage{aliascnt}
\usepackage{etoolbox}
\usepackage{longtable}
\usepackage{tabu}
\usepackage{float}
\usepackage{amssymb}
\usepackage{enumitem}
\usepackage{graphicx}
\usepackage{pgfplots}
\usepackage{etoolbox}
\usepackage{fancyhdr} % to change header and footers
\usepackage{blindtext} 
\usepackage{tabularx} 
\usepackage{graphicx}



%Now all the Spacing for Sections




\frontmatter
 
%

%

\newcolumntype{b}{X}
%
\setcounter{page}{0}
\title{Pricing of Cloud Services}
\author{Sandro Luck 13-927-769}

\institute{Seminar in Advanced Software Engineering, HS 16
\\
\today
}
%

\begin{document} 

\let\oldaddcontentsline\addcontentsline
\def\addcontentsline#1#2#3{}
\maketitle
\def\addcontentsline#1#2#3{\oldaddcontentsline{#1}{#2}{#3}}
\fancyhead{}
\fancyfoot{}
\fancyhead[R]{\thepage}


% for the preliminaries
%

\tableofcontents

\newpage
%\begin{minipage}{\textwidth}

\pagenumbering{arabic}
\begin{abstract}
Cloud Services are seen as an ever-growing new business field. They give businesses around the world the possibility to reduce costs and increased their Quality of Service while making a healthy profit. Different providers are competing for the market and each with their individual pricing, software and hardware approach. 
Every public cloud customer is looking to find the provider offering the lowest price and the best Quality of Service. To understand what the lowest price and the best Quality of Service is one has to understand what the Total Cost of Ownership of a cloud environment is, and how the upper and lower limits of prices are formed. The purpose of this paper is to discuss how prices in this area are formed and what aspects must be considered when discussing the price differences from up to 1000\% of cloud instances.
    \keywords{cloud,  pricing scheme, data center, cloud provider}

\end{abstract}

\section{Introduction}\label{intro}

\setcounter{page}{1}
In recent years, IT cloud services grew with double-digit rates~\cite{statista}. 
It has been found that 17\% of enterprises in the IT-field now have more than 1000 VMs in public clouds, and 31\% in private clouds~\cite{weins}. 
While potential margins have been estimated to be as high as 90\% in earlier years, the most recent numbers from Amazon suggest the current margins are at around 23.5\%~\cite{darrow}, which is still more than the most profitable industry's average profit margin in the U.S. (around 21\% in the health technology area~\cite{forbes2}).
\newline
Cloud computing is seen as a heavily growing and fast adapting business which has the potential to lower costs, increase flexibility and balance server loads. More and more providers are gaining foothold in this market and each with his own variation in price, functionality and hardware.\\ 
Cloud Computing refers to services, hardware and system software provided via data centers~\cite{armbrust}. It can be differentiated between private, public and hybrid clouds. While a public cloud is designed for open use by the general public, a private cloud is used exclusively by a single organization and hybrid clouds can be imagined as a composition of both of these~\cite{mell}. The main focus of this paper will lie on public clouds, even though some aspects might also apply to private clouds. \\
Also, the Service Models offered by the cloud providers can be categorised into Software as a Service (SaaS), Platform as a Service (PaaS) and Infrastructure as a Service (IaaS)~\cite{mell}. While SaaS provides the consumer with an application hosted on a cloud, usually owned by the provider, PaaS and IaaS give him the ability to deploy and run his own application on the provider's systems~\cite{mell}. PaaS and IaaS further differ mainly in the area that in PaaS
\begin{displayquote}
" The consumer does not manage or control the underlying cloud infrastructure"~\cite{mell}
\end{displayquote}
while with IaaS the customer does. It is worth noting that often web based solutions are advertised as cloud computing solutions, also known as "cloud washing"~\cite{ercolani}, which might further complicate the terminology.\\
In the following sections, different pricing aspects from different perspectives (customer and provider) will be discussed to understand how the lower and upper limits of prices are formed and in what form they are presented. To understand the upper limit of prices a look at the Total Cost of Ownership of a cloud environment will be taken in section \ref{customer}, which determines the upper limit of prices the customers are willing to pay. To understand the lower limit of prices providers are willing to accept, the main cost factor all cloud providers have in will be discussed in section \ref{datacenter}, which are mainly related to data centers. To further understand how to compare and evaluate possible services we will look at the presentation form of these prices (e.g. pricing schemes) in section \ref{scheme}.
%
\section{Customer Cost Types}\label{customer}

As described in section~\ref{intro}, this sections serves the purpose of understanding the upper limit a customer is willing to pay for a cloud solution and therefor influences the pricing. We assume that the upper limit a customer is willing to pay for a public cloud solution is the amount he is expecting to pay for a private cloud solution of equal functionality and capability.
When it comes to determining how much setting up and maintaining a cloud services costs, several factors must be taken into account, both time and monetary investments will be discussed to understand the Total Cost of Ownership of a cloud solution. The research group in Accounting and Information Systems from the University of Osnabr{\"u}ck identified several cost factors, a potential first setup of a cloud service might generate~\cite{martens}. Several of these will be discussed in the following descriptions, which will mainly focus on the customer's perspective, an overview is given in table~\ref{table:TCO}.

\begin{center}
\resizebox{0.69\textwidth}{!}{\begin{minipage}{\textwidth}
\begin{table}[H]
\caption{Total Cost of Ownership Cloud}
\begin{tabularx}{\textwidth}{|p{3.5cm}|b|}
            \hline
            \textbf{Cost Type} & \textbf{Cost Factors}  \\ 
            \hline \hline
            Strategic Decision & expenditure of time, consulting services, information for decision-making~\cite{martens}\\ \hline
            Evaluation and Selection of Service Provider & expenditure of time, consulting services, information for decision-making~\cite{martens}\\ \hline
            Service Charge & 
            user-dependent basic charges, storage
capacity [for the developer team],
inbound data transfer, outbound data
transfer, provider internal data transfer, extra user data storage capacity,
extra user document storage capacity,
queries to the Application Programming
Interface, sent emails, database, secured logins, connections with
other providers’ applications, number of queries, domain, SSL certificate, access to the service system, user, computing power~\cite{martens} \\ \hline
Implementation,
Configuration, 
Integration and
Migration & expenditure of time, porting process~\cite{martens} \\ \hline
Support & expenditure of time, support costs,
problem solving~\cite{martens} \\ \hline
Initial and permanent
training & preparation time of internal employees,
participating time of internal employees,
instruction material, external consulting
services~\cite{martens} \\ \hline
Maintenance and
Modification & expediture of time~\cite{martens} \\ \hline
System Failure & loss per period~\cite{martens} \\ \hline
Backsourcing or
Discarding  & expenditure of time, porting process~\cite{martens} \\ \hline
\end{tabularx}
\label{table:TCO}
\end{table}
\end{minipage}}
\end{center}
    %TABLE

    
    
    
    
\subsection{Strategic Decision}
Strategic Decision Costs include all costs which are necessary to select a suitable service and cloud type. It is the sum of expenditure of time needed for decision making, expenses for information (e.g. subscription to statistics provider) and potential costs for external consulting services~\cite{martens}. For smaller projects this costs might be negligible, but for bigger projects a lot of effort has to be put into making these decisions. Depending on the varying needs for big scale projects this might include concerns about essential characteristics as on-demand self-service, rapid adjustment to elasticity and security concerns, to only name a few related aspects~\cite{ercolani}.

\subsection{Evaluation and Selection of Service Provider}
The evaluation and selection of a specific service provider are a cost type which include the cost factors expenditure of time, potential consulting services and information for decision-making (e.g. books)~\cite{martens}. General aspects are finding the providers which offer the desired service evaluation of the functionality and the providers Service-level agreement (SLA). After the basic needs for the project have been found, an analysis which identifies a specific provider can be made. Such an analysis might look at factors such as penalty on early termination, data return on subscription cancel, provider user training, training charge fee, self-support/documentation, customer support by phone, email or ticket, manageable transferability of data, authentication and the location of the information country to only name the most weighted key features Ercolani G. identified~\cite{ercolani}. Also Martens B. et al. found that
\begin{displayquote}
"The average reputation of the
Cloud Computing service provider and the use of standard data
formats are more important than financial aspects."~\cite{martens2} 
\end{displayquote}
so also, the company behind the service will be evaluated. In fact, Armbrust M. et al. found that 
\begin{displayquote}
"Some Internet service providers today
cost a factor of 10 more than others
because they are more dependable and
offer extra services to improve usability."~\cite{armbrust}.
\end{displayquote}
This makes selecting a cloud provider both an important and complex decision.

\subsection{Service Charge}
Service Charge is determined by the pricing scheme, which will be further discussed in section \ref{scheme}, the provider uses to bill the subscriber~\cite{martens}. These pricing schemes vary heavily and may range from auction based systems, like Amazon's EC2 spot instance pricing mechanism, which sold unused server capacities on an, advertised as free auction based market~\cite{ben}, to traditional flat rate based subscriptions~\cite{dincloud}. Different cloud providers choose different approaches to create their unique Service Charge. Most features a cloud service can offer might be a billing position for certain cloud providers, e.g. SSL-certificates are a very popular additional feature which can be billed~\cite{apurva}. The Service Charges vary based on whether they represent a SaaS, IaaS or a PaaS service.
\subsubsection{SaaS}
The simplest Service Charge to calculate is the one for SaaS. It is mainly determined by licensing costs for a software, usually the customer may choose from several packages (e.g. basic package, premium package) and is mostly a fixed price per license, thus the price a company pays is the number of users multiplied with the cost per license~\cite{martens}. In some cases, licenses for entire organisations are purchasable. Different billing models exist including free, per month, pay-as-you-go, per login, pay-as-you-use and per item (e.g. number of documents on docverse.com)~\cite{bibi}. Several of these schemes will be looked at in section~\ref{scheme}.
\subsubsection{IaaS}
The cost factors associated with IaaS are mainly hardware and connection dependent. Its price is determined by the computing power, inbound data transfer, outbound data transfer, provider internal data transfer, number of queries, domain, SSL certificate, licence and the basic service charge~\cite{martens}.  The hardware dependent cost factors tend to lower with technical advancement. Data transfer costs are of special interest since the advancement of the Internet of Thing (IOT) will increase the traffic~\cite{gartner} and minimizing this traffic will become a crucial optimization. Especially keep-alive messages are interesting in this area since they account for about 45\% of the total costs under a typical remote-control workload~\cite{daisuke}.
\subsubsection{PaaS}
Platform as a Service pricing schemes are arguably the most complex to evaluate since they usually represent an entire ecosystem. The cost factors include user-dependent basic charges, storage capacity (for the developer team), inbound data transfer, outbound data transfer, provider internal data transfer, extra user data storage capacity, queries for the Application Programming Interface, sent emails, database, secure logins and connections with other provider's applications~\cite{martens}. Programming on a specific platform (e.g. Heroku) might reduce the complexity related to the infrastructure, but could also increase the risk of "Data Lock-In". A way to avoid increasing dependency on one provider is to use standardized APIs and the potential use of hybrid clouds~\cite{armbrust}.

\subsection{Implementation, Configuration, Integration and Migration}
Migration, Integration and Implementation costs might vary heavily based on the differences between the old system and the new one. While the configuration costs vary based on the provider and the applications needs. The implementation and integration costs are traditionally the highest since a lot of effort has to be put into getting it up and running. Miscalculations in these areas are consequently especially painful, while balancing a budget, and surveys suggest that most projects (60\%) encounter these kinds of problems~\cite{molokken}. 

\subsection{Support}
The term support will be uses as everything the provider might do to help the customer with the service provided by the provider (e.g. the cloud service). Support might include email, phone, ticket support or text chats~\cite{martens}. Depending on the pricing scheme, support is either free or billed. While for SaaS support is often a minor cost factor. With IaaS and PaaS, a good support is crucial and accordingly often included in premium packages. Some providers offer some interaction with the support for free, mainly to help with the evaluation of the products and clarify functionality and services offered by the provider~\cite{ercolani} (e.g. sales and marketing related tasks).

\subsection{Initial and permanent training}
Initial and permanent training costs are the costs which occur when training the companies employees. Mainly the time the employees spend while learning new technologies~\cite{martens}. It summarises the preparation time internal employees might need to prepare workshops or similar training events or external consulting services which might charge for the same. The time all participants of such workshops might invest~\cite{martens} should also be considered and often instruction material (e.g. books or software). Key features are the similarity to the technologies already used, ease to setup, speed - implementation time, customization, configurability, potential new security barriers and the ease of porting data to the new system~\cite{ercolani}.

\subsection{Maintenance and Modification}
Maintenance and Modification are used in the context of time employees have to spend to maintain or modify the existing implementation~\cite{martens}. This cost type increases the longer the service is up, since the provider tends to add new features the customer's software has not fully utilized (e.g. new feature which make the service more efficient). Changes in API, especially deprecation increases this costs, the impact is determined by the frequency of changes, the magnitude of the changes and consistency of adaptation (e.g. is there a single go-to solution or only a workaround)~\cite{romain}.

\subsection{System Failure}
While system failures are not a direct cost, often the customer even gets a voucher or reduction on his subscription, they have a heavy impact on business processes in general. The consequences of a system failure strongly depend on the services hosted~\cite{martens}, when considering a bank a system failure could  produce losses in the millions, while with less critical task the loss might be neglectable~\cite{evolven}. Lost working time, contract penalty for non-delivery of services and even worse a loss of reputation is only some of the possible problems~\cite{martens}. Another potential problem that arises is reputation fate sharing (e.g. illegal activities of others using the same cloud might affect your reputation) which might lead to blacklisting of IP addresses~\cite{armbrust} and can be described as an indirect system failure.

\subsection{Backsourcing or Discarding}
In recent years, the importance of (big-)data has increased and some cloud providers even launched special platforms for processing these~\cite{hadoop}. The Backsourcing or Discarding cost type includes the costs of porting data to or from the system~\cite{ercolani} and are influenced by several factors. The data size and similarity of format mainly influence the time an employee might have to port the data~\cite{ercolani,martens} and potential security concerns could have an impact as well~\cite{ercolani}. Also, some legal concerns regarding the data ownership have to be taken into account~\cite{ercolani}.

\section{Cost Types Data Center}\label{datacenter}
Potential costs a cloud environment generates, from an users perspective, have been described. While a lot of information can be found on how to determine and reduce these costs, information on the cost factors of cloud providers are rare but crucial to understand the base costs of cloud services. These costs are the lower limit of prices providers can afford in the long run. 
The cost type all cloud service providers share come from data centers and are mainly determined by two cost types, acquisition costs and administration costs~\cite{smith}. These costs influence the lower bound of Cloud Service prices~\cite{sharma}. The term 'data center' refers to facilities to house computer systems and associated components~\cite{maciel}. The acquisition costs are the cost the providers spend on property and equipment. The costs that arise with developing software will not be looked at since they vary heavily based on the used cloud (e.g. IaaS vs SaaS). The administration costs relate to energy consumption, employees and infrastructure and maintaining costs in general. Three subsystems, described by Maciel P. et al. will be considered, the IT infrastructure, the power infrastructure and the cooling infrastructure. Also, the main advantages of data centres will be discussed the possibility to us economy of scale related techniques to lower the acquisition and administration costs.

\subsection{Acquisition Costs}
The IT infrastructure consists of servers, networking equipment and storage devices~\cite{maciel}. The cooling infrastructure includes cooling
towers, chillers and CRAC units~\cite{maciel}. The power infrastructure is responsible for supplying the cooling infrastructure and IT infrastructure with a reliable source of power. The consumption of all U.S. data centers was 17000 MW in 2013 and is estimated to grow to 25500 MW in 2020~\cite{thibodeau}. The power infrastructure includes AC sources,
low voltage panels, uninterruptible power supplies (UPS), transformers, static transfer switches,
subpanels and junction boxes~\cite{maciel}. The costs Koomey J. found suggest that about 79.3\% of the IT capital costs are server costs 15.5\% are storage device costs and 3.5\% are networking costs~\cite{koomey}. Of the kW related infrastructure costs are 90.3\% server related, 5.0\% storage device related and 4.4\% network related. Of the other capital costs the total costs per rack are 80\% server related costs, 10\% are storage cost (about 8.3\% for disk storage and 1.7\% are Tape storage costs) and 10\% are networking costs~\cite{koomey}. Koomey found the total installed capital costs to be 5091 \$/square feet of electrically active floor space (approximately 54800 \$/$m^2$ of electrically active floor space). This results in annualized capital costs of 19.1 Million \$ for a data center with 5120 units of servers. As shown in figure~\ref{fig:ancap} of these capital costs approximately 53\% are kW related infrastructure costs, about 29\% are the total IT costs (e.g. servers), around 10.3\% are other costs (e.g. architectural and engineering costs, gas and fire suppression etc.), 4.1\% are related to the interest during construction, 3.6\% are Point of Presence (POP) costs while Land costs surprisingly account for less than 1\% of the capital costs.
\
\begin{center}
\begin{figure}
    \caption{Annualized capital costs data center}
\begin{tikzpicture}
    \pie[text=legend,radius=2.5]{53/kw related infrastructure costs, 29/Total IT costs, 10.3/Other costs, 4.1/Interest during construction, 3.6/POP}
\end{tikzpicture}
        \label{fig:ancap}

\end{figure}
\end{center}



\subsection{Administration Costs}
While the annualized capital costs are about 76\% of the total annualized costs, the total operating costs per year make up for around 24\% of the total costs per year~\cite{koomey}. As seen in figure~\ref{fig:oper} the total operating costs per year are mainly determined by the 42\% for electricity costs, but also about 15.4\% for management staff (IT-management and facility site management) around 11\% for Security and property taxes each, about 10\% for costs related to maintenance, 8\% for networking fees and 3\% for janitorial services~\cite{koomey}. Cost for the development of additional software, which in case of SaaS form a major cost factor, are not considered here. The named numbers only relate to the costs, generate by a data center.
\begin{center}
\begin{figure}
    \caption{Operating costs data center}
\begin{tikzpicture}
    \pie[text=legend, radius=2.5]{42/Electricity costs, 15.4/Management staff, 11/Security, 11/Property taxes,10/Maintenance, 8/Networking fees,3/Janitorial }
\end{tikzpicture}
        \label{fig:oper}

\end{figure}
\end{center}

\subsection{Cost effects}
They named cost factors can additional vary based on several effects. To further explain why the pricing difference between different cloud providers might reach a factor 10~\cite{armbrust} we will have a look at several effects. Sharma B. et al. found several effects which influence the resource price namely the effect of initial investment, the effect of contract period, the effect of rate of deprecation, the effect of Quality of Service (QOS) and the effect of resource age~\cite{sharma}. They found that the resource price increases proportional to the initial investment. The effect of QOS describes that the resource price increases proportionally to the increase in QOS~\cite{sharma}. The effect of contract time on cloud resource describes that the price decreases as the contract time increases. They found that this is due to two reasons, the resource price variation could average out over a period of time and smaller jobs may be executed at time where resource prices are low~\cite{sharma} (e.g. electricity at night). The effect of age of resource is assumed to have no effect on the resource price since the client is concerned in the quality of work and not the hardware used to accomplish the task~\cite{sharma}.

\subsection{Economy of scale}
The main advantages of public clouds, given a reasonable size, compared to private cloud solutions can be summarised under the term economy of scale. The term economy of scale will be used as a proportionate saving in costs gained by an increased level of production (e.g. more servers and customers). Brynjolfsson E. et al. described that the economy of scale related factors of cloud computing are similar to those in electricity generation~\cite{brynjolfsson}. They both, the cloud computing and the electricity generation technologies, can be seen as general-purpose technologies 
\begin{displayquote}
"General-purpose technologies, or GPTs, are best thought of not as
discrete tools but as platforms on which
many different tools, or applications,
can be constructed."~\cite{brynjolfsson}
\end{displayquote}
 And GPTs can be assumed to replace the private cloud providers, given time and the possibility to provide the technologies centrally~\cite{brynjolfsson}. But they also named that in case of electricity this improvement process took 30 to 40 years to bring the full benefits and it can be assumed that it will take several years until the IT industry fully utilizes the potential savings. Some of the advantages include statistical multiplexing, centralized infrastructure, less system administrators and also placing data centers closer to power production can generate savings, since the costs of transferring data is lower than the costs of transferring energy~\cite{brynjolfsson}. Apart from the savings brought by energy which accounts for approximately 10.1\% of the annualized total costs~\cite{koomey} some benefits related to economy of scale can reduce the acquisition costs~\cite{brynjolfsson} in other words, if one buys a CPU for 100\$ a data center might buy 8 for 700\$. The same principle holds for most equipment and can reduce the total annualized acquisition costs.
 If demand for a service varies with time, like it usually does with cloud computing, it is beneficial for the actors involved to balance their loads evenly. Elasticity describes the ability to adapt to the current work load. If the provider uses a 100\% of his resources during peak load (the time the data center runs at maximum capacity) he might have an high average utilization of 80\% while a private cloud has a lower average utilization since he is only running a few applications. A simple example might be a Christmas shop having peak loads in November and December while only using a fractions of this computational power during the other 10 months of the year. In such a case a public cloud provider hosting a variety of service (e.g. Easter shops, Valentines day, Halloween) can be assumed to  have a higher average utilization.
 
\section{Pricing Schemes}\label{scheme}
In the preceding sections the cost factors of both the customer and the cloud service provider have been described. In the section~\ref{customer} the most important cost types from a customer's perspective have been discussed, these cost type describe the Total Cost of Ownership (TCO) setting up a cloud service and maintaining it will have. These costs combined form the price the customer has to pay for a public cloud service. As long as, this price is lower as the TCO the customer would be expected to have, when running a private cloud (these costs will be referred to as private costs), renting a public cloud is beneficial from a customers perspective. The section~\ref{datacenter} described the most important costs a cloud provider hosting a data center has. These costs, broken down to the individual resources to rent away for profit, define the lower bound of the price the provider is willing to accept for his service (these costs will be referred to as provider's costs). As long as the customer's private costs are higher than the provider's public costs a cooperation is beneficial for both parties.\\
Different factors which influence the price have been identified, but the presentation form, the way the individual QOS aspects are priced, has a strong influence on the final amount the customer is priced. Mazrekaj A et al. have collected several pricing models, both existing and theoretical (pricing scheme not yet implemented), and compared them. Several of them will be described and compared in the following subsections to give an overview of possible pricing schemes.
\subsection{Pay-as-you-go}
Pay-as-you-go is a constant price set by the provider and can be seen as a static model~\cite{artan}. Pay-as-you-go platforms charge by the resources used by the customer (e.g. CPU, storage, OS, security)~\cite{margaret} rather than to pay for a fixed amount of resource each month. Pay-as-you-go has been known and used as a pricing scheme for over 100 years as a tax system~\cite{payghistory} and has been popular with mobile phone billing (e.g. prepaid mobile phone billing). This is probably the most common pricing scheme used (with IaaS and PaaS) and is often advertised as a fair and easy to understand pricing model. Microsoft's Azure uses pay-as-you-go and advertises it as 
\begin{displayquote}
"No commitment. Pay for what you use each month"~\cite{azureprice}.
\end{displayquote}
Also, Amazons AWS pricing is pay-as-you-go based and advertised similarly as
\begin{displayquote}
"Pay-as-you-go pricing allows you to easily adapt to changing business needs without over committing budgets and improving your responsiveness to changes"~\cite{awsprice}.
\end{displayquote}
As these two quotes suggest, pay-as-you-go is advertised as a flexible and fair model for cloud pricing. Two aspects which seem to be specially appealing to cloud customers since they also are some of the main advantages of public over private clouds.
However, it has been described as a potentially unfair pricing scheme, mainly because of its traditionally static implementation since the customer might pay more than necessary~\cite{artan} (e.g. electricity at night might be cheaper). Well known providers using pay-as-you-go include Amazon EC2~\cite{margaret}, Microsoft's Azure~\cite{azureprice} and IBM's XIV Storage System.
\subsection{Subscription}
Subscription based pricing models is a static model and bills the customer a fixed amount per subscription~\cite{artan}. Subscription based pricing is one of the most common seen among SaaS. Prices can change and providers might charge more or less~\cite{artan}. Usually the price is paid monthly and per user, but specially for business software it is common to charge per company. Subscription based pricing has been known and used for a long time outside of the cloud industry (e.g. gym subscription). Often services using this billing system have several subscription plans with different service-level agreements and different prices. These subscription plans mostly range from a basic package (cheapest, least functionality) to premium package (most expensive, most functionality), note that the naming of the subscription plans often varies (e.g. premium/pro/business package, and basic/personal/essential package). Often these services also offer a free trial period or the basic package is free with an option to upgrade the service. This is mainly due to marketing reasons~\cite{marketing}. Some well-known examples include Microsoft's OneDrive~\cite{onedrive}, Dropbox~\cite{dropbox} and Spotify~\cite{spotify}.
\subsection{Pay-for-resources}
This is a static model and offers a maximum utilization of resources~\cite{artan}. A pricing scheme very similar to the pay-as-you-go scheme we discussed earlier. With pay-for-resources the customer defines an amount of resources he is willing to use and pay and is billed this amount. The customer defines the resources used either explicitly (explicitly defining the resources to be used a month upfront) or implicitly (using the resources and paying for them later). Pay-as-you-go is often a special case of pay-for-resources. So, the price is the sum of the individual resources rented and again mostly is billed monthly. What features or infrastructures are seen as a resource, and accordingly billed, varies looking at different provider. It has been described as a fair pricing model for both providers and customers even though it might be hard to implement due to the varies problems when calculating the monetary values of individual resources~\cite{artan}. The pay-for-resources pricing scheme is not often seen in a pure form, they are combined with other billing possibilities. A well-known example might be Google's App Engine, but not in it's pure form since they combine this with several different pricing schemes.


\subsection{Hybrid pricing}
This pricing scheme is both static (price limits are static) and dynamic (prices are adjusted dynamically), has a low computational overhead and is seen as fair to customers~\cite{may}. One obstacle is that the common base prices and variation limit must be defined. In this model, proposed by Intel, prices are determined by the waiting time in the job queue~\cite{artan}, thus vary with supply and demand. The goal of this system was to implement an automated system which adapts to rapid changes in demand~\cite{intel}. They implemented a system which can port application from one cloud to another to be able to switch between different cloud providers fast and without too much effort, to outsource some of the computations to public clouds and some to their private cloud.  Intel used this approach to avoid building excessive private clouds and scale their project, while developing, to meet their user's needs until they can predict the server loads and implement a stable private cloud~\cite{intel}.

\subsection{Dynamic resource pricing}
Dynamic Resource Pricing is a theoretical model. The resources are priced based on demand and supply using distributed auctions~\cite{artan}. It is a strategy-proof dynamic pricing model used for federated clouds to allocate shared resources with multiple resources types~\cite{meng}. Strategyproofness is a term in game theory to describe a state in which it is a weakly-dominate strategy to reveal private information (e.g. the price he is willing to pay).
\begin{displayquote}
"A federated cloud (also called cloud federation) is the deployment and management of multiple external and internal cloud computing services to match business needs"~\cite{techtarget}.
\end{displayquote}
In other words a customer uses internal and external clouds simultaneously. This approach is very similar to a hybrid cloud~\cite{stackoverflow}. The actual pricing is set via distributed auctions and has been proposed to be implemented as a peer to peer network to improve scalability. It has been found to increase user's welfare and increases the amount of successful buying requests under high loads but decreases it under lower load, compared to a fixed pricing system~\cite{meng}. An example for a federated cloud would be ownCloud which offers federated cloud sharing, a service which simplifies the sharing of documents with different security levels~\cite{owncloud}. 

\subsection{Value-based
pricing}
This model has been implemented and the prices are formed in a dynamical way. Prices are set according to the value perceived by the customer~\cite{may} (what he is willing to pay). From a provider perspective, this approach generates high revenues but it might be difficult to obtain the data necessary to determine the price~\cite{may}. Value based pricing has been known and used in different fields including the software industry and the pharmaceutical industry. Both industries tend to have products which have a small production costs but high initial development costs. In such an industry value-based pricing can be used as a guide line to adjust prices. Thus, increasing the profit the provider can realise without charging more than the customer is willing to pay.

\subsection{Cost-based pricing}
This is a dynamic model which has been implemented and it is relatively easy to calculate the price for it~\cite{may}. The prices are formed by looking at the provider's costs (total costs of ownership) and adding a profit on top. This model guaranties the provider, given enough customers willing to pay, that he will always make a profit. Compared to other pricing model the data necessary to calculate the price is easy to obtain, mainly because all the providers cost data should be available without further evaluations. 
To calculate this price one only needs the costs, the number of customers and the profit margin the provider is planning to realise. It has been described as unfair to customers since it tends to ignore his role entirely~\cite{may}. This pricing system has been known for a very long time and still has an influence on modern pricing, however it is mostly used in industries outside the cloud environment.

\subsection{Competition-based pricing}
Competition-based pricing is a dynamic pricing model which has been implemented and plays a role in most pricing decision~\cite{may}. The price is formed by looking at the prices of other competitors and  changing its own price purposely based on their prices~\cite{may}. Since we change our prices in this approach based on dynamic prices of other competitors, our model is dynamic as well. Prices formed in this way may go below or above the values set by the competition, depending on other aspects such as the amount and quality of our features compared to our competitions features. Calculating this price is relatively easy since the information (the competitor's prices), necessary to make the decision, is mostly publicly available. This model tends to ignore the customer's role entirely but has been described as fair to customers, since through the competition his wishes are taken into account indirectly~\cite{may}.

\subsection{Customerbased pricing}
In this dynamic approach the price is set to the amount the customer is prepared to pay for the respective service~\cite{may}. Since what the customer is prepared to pay often varies this is a dynamic approach, which changes with the customers financial situation and his perception of the service. The price is formed by finding the amount a customer is prepared to pay for a service and charging this amount. This guarantees that the customer is always willing to pay the providers prices, but since the data necessary to make that decision is very hard to obtain, implementing it is rather difficult~\cite{may}. This is because customers rarely indicate to the provider what they are willing to pay. Possible ways to obtain this data, like surveys or interviews, often lack absolute certainty, since the customer might indicate a different price than the price he is prepared to pay in order to manipulate the decision. It is seen as a fair approach to customer and takes the customer role always into account~\cite{may}.

\subsection{Double sided Combinational Auctions to Resource Allocation}
In this model the providers and users deal through double-sided combinational auctions~\cite{artan}. In these kind of auctions buyers and sellers can place orders on combinations of different securities~\cite{schellhorn}. A buyer would then specify how much for each resource he is willing to pay, which resource and how much of each individual attribute he is willing to buy. In the end the best matches for prices and attributes will be matched~\cite{schellhorn}. These orders are conjunctive so the orders are only matched if the full bundle is available~\cite{schellhorn}. This approach is fair to the customer and the provider since their bid is matched to a provider or customer which has a similar perception of what the service is worth. Double sided combinational auctions is a theoretical model which has been described but not implemented yet.

%\subsubsection{Pricing algorithm for cloud computing resources}
%This is a theoretic real-time dynamic pricing method which aims at reducing the costs and maximizing the revenues for providers~\cite{may}. Compared to other similar approaches it aims mainly at improving the economical aspects. A possible critic is that its probably not suited for rapid changes in supply and demand~\cite{may}.
\section{Conclusion}
We have seen several different factors which must considered when looking at how prices in the cloud area are formed. Both the customers cost factors and the providers main expense have been discussed to understand the upper and lower limit of modern cloud pricing. Different presentation forms for prices, the pricing schemes, have been discussed to understand how prices are formed currently and how they might change in the future. The cloud computing business shows several parallels to the electricity industry in its beginning. And as with electricity the full potential of public clouds has not yet been unleashed. Assuming the cloud environment develops as the electricity industry we can expect some changes in the near future which will make this technology even more accessible and affordable. Both improving pricing schemes and data centers might make this happen. 

\newpage

% ---- Bibliography ---- 
%
\begin{thebibliography}{}
%
\bibitem[1]{statista}
Statista.:
\textit{Market growth forecast for public IT cloud services worldwide from 2011 to 201 (2016)}
\\\texttt{ www.statista.com/statistics/203578/global-forecast-of-cloud-computing\\-services-growth/}

\bibitem[2]{weins}
Weins, K.:
\textit{Cloud Computing Trends: 2016 State of the Cloud Survey (2016)}
\\\texttt{ www.rightscale.com/blog/cloud-industry-insights/cloud-computing-trends\\-2016-state-cloud-survey}

\bibitem[3]{darrow}
Darrow, B.:
\textit{Amazon Web Services Forges Ahead Like There Is No Competition (2016)}
\\\texttt{ fortune.com/2016/04/28/amazon-web-services-forges-ahead-like-there-is-\\no-competition/}

\bibitem[4]{armbrust}
Armbrust M., Fox A., Griffith R., Anthony J., Katz R., Konwinski A., Lee G., Patterson D., Rabkin A., Stoica I., Zaharia M.:
\textit{A View of Cloud Computing (2010)}
\\\texttt{ dl.acm.org/citation.cfm?id=1721672}

\bibitem[5]{mell}
Mell P., Grance T.:
\textit{The NIST Definition of Cloud Computing (2011)}
\\\texttt{ faculty.winthrop.edu/domanm/csci411/Handouts/NIST.pdf}

\bibitem[6]{ercolani}
Ercolani G.:
\textit{Cloud Computing Services Potential Analysis (2013)}
\\\texttt{ dspace.unitus.it/handle/2067/2356}

\bibitem[7]{dincloud}
Dincloud:
\\\texttt{ www.dincloud.com/Pricing (2016)}

\bibitem[8]{apurva}
Apurva, J.:
\textit{Buy and Configure an SSL Certificate for your Azure App Service (2016)}
\\\texttt{ azure.microsoft.com/en-us/documentation/articles/web-sites-purchase-\\ssl-web-site/}

\bibitem[9]{molokken}
Molokken K., Jorgensen M.:
\textit{A Review of Surveys on Software Effort Estimation (2003)}
\\\texttt{ ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=1237981}

\bibitem[10]{ben}
Ben-Yehuda O., Ben-Yehuda M., Schuster A., Tsafrir D.:
\textit{Deconstructing Amazon EC2 Spot Instance Pricin (2013)}
\\\texttt{ dl.acm.org/citation.cfm?doid=2509413.2509416}

\bibitem[11]{martens}
Benedikt Martens, Marc Walterbusch and Frank Teuteberg:
\textit{Costing of Cloud Computing Services: A Total Cost of Ownership Approach (2012)}
\\\texttt{ ieeexplore.ieee.org/document/6149074/?arnumber=6149074}

\bibitem[12]{romain}
Romain Robbes, Mircea Lungu, David Röthlisberger:
\textit{How Do Developers React to API Deprecation? The Case
of a Smalltalk Ecosystem{2012}}
\\\texttt{ users.dcc.uchile.cl/~rrobbes/p/FSE2012-deprecation.pdf}

\bibitem[13]{evolven}
Alex Gutman and Martin Perlin:
\textit{10 Devastating Outages and Failures of Major Brands in 201 (2012)}
\\\texttt{ www.evolven.com/blog/2011-devastating-outages-major-brands.html}

\bibitem[14]{hadoop}
Microsoft:
\textit{Hadoop. What is Hadoop (2016)}
\\\texttt{ azure.microsoft.com/en-us/solutions/hadoop/}

\bibitem[15]{bibi}
Stamatia Bibi, Dimitrios Katsaros, Panayiotis Bozanis:
\textit{Business application acquisition:
On-premise or SaaS-based solutions (2012)}
\\\texttt{ ieeexplore.ieee.org/document/6188600/}

\bibitem[16]{gartner}
Gartner:
\textit{Gartner Says the Internet of Things Will Transform the Data Cente (2014)}
\\\texttt{ www.gartner.com/newsroom/id/2684616}

\bibitem[17]{daisuke}
Daisuke Ajitomi, Hiroshi Kawazoe, Keisuke Minami, Naoki Esaka:
\textit{A Cost-Effective Method to Keep Availability of Many Cloud-connected Device (2015)}
\\\texttt{ ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7214021}

\bibitem[18]{martens2}
Benedikt Martens, Jens Poeppelbuss and Frank Teuteberg:
\textit{Understanding the Cloud Computing Ecosystem:
Results from a Quantitative Content Analysi (2011)}
\\\texttt{ pdfs.semanticscholar.org/e029/40169ee61d270c4b1912be23b1801c22268a.pdf}

\bibitem[19]{sharma}
Bhanu Sharma, Ruppa K. Thulasiram, Parimala Thulasiraman, Saurabh K. Garg	:
\textit{Pricing Cloud Compute Commodities: A Novel
Financial Economic Mode (2012)}
\\\texttt{ www.cloudbus.org/papers/PricingCloud-CCGrid2012.pdf}

\bibitem[20]{maciel}
J. Figueiredo, P. Maciel, G. Callou, E. Tavares, E. Sousa, B. Silva:
\textit{Estimating Reliability Importance and Total Cost
of Acquisition for Data Center Power
Infrastructure (2011)}
\\\texttt{ ieeexplore.ieee.org/document/6083702/}

\bibitem[21]{koomey}
Koomey J.:
\textit{A Simple Model for Determining True Total Cost of Ownership for Data Center (2007)}
\texttt{www.researchgate.net/publication/252096634\_A\_Simple\_Model\_for\_Determining\\\_True\_Total\_Cost\_of\_Ownership\_for\_Data\_Centers}

\bibitem[22]{thibodeau}
Patrick Thibodeau  :
\textit{data centres are the new polluters
(2014)}
\\\texttt{ www.computerworld.com/article/2598562/data-center/data-centers-are-the\\-new-polluters.html}

\bibitem[23]{brynjolfsson}
Erik Brynjolfsson, Paul Hofmann, and John Jordan:
\textit{economic and
Business dimensions
Cloud Computing and electricity:
Beyond the Utility Model
(2010)}
\\\texttt{ citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.188.1662}

\bibitem[24]{smith}
Julie Smith David, David Schuff and Robert St. Louis:
\textit{Managing your total IT cost of ownership}
\\\texttt{ dl.acm.org/citation.cfm?id=502273}

\bibitem[25]{artan}
Artan Mazreka, Isak Shabani, Besmir Sejdiu:
\textit{Pricing Schemes in Cloud Computing: An Overview
}
\texttt{thesai.org/Downloads/Volume7No2/Paper\_11-Pricing\_Schemes\_in\_Cloud\_Computing\\\_An\_Overview.pdf}

\bibitem[26]{margaret}
Margaret Rouse:
\textit{Pricing Schemes in Cloud Computing: An Overview
}
\verb|pay-as-you-go cloud computing (PAYG cloud computing)|

\bibitem[27]{techtarget}
Margaret Rouse:
\textit{federated cloud (cloud federation)}
\\\texttt{ whatis.techtarget.com/definition/federated-cloud-cloud-federation}

\bibitem[28]{stackoverflow}
Matt Smith (stackoverflow user name):
\textit{Hybrid Cloud vs Federated Cloud top answer}
\\\texttt{ stackoverflow.com/questions/12988988/hybrid-cloud-vs-federated-cloud}

\bibitem[29]{owncloud}
OwnCloud:
\textit{Federated Cloud Sharing in Action}
\\\texttt{ owncloud.com/federated-cloud-sharing/}

\bibitem[30]{may}
May Al-Roomi, Shaikha Al-Ebrahim, Sabika Buqrais and Imtiaz Ahmad
:
\textit{Cloud Computing Pricing Models: A Survey
}
\\\texttt{ www.sersc.org/journals/IJGDC/vol6\_ no5/9.pdf}

\bibitem[31]{intel}
Intel :
\textit{Cloud Computing Cost: Saving with a Hybrid Model
}
\\\texttt{http://www.intel.com/content/dam/www/public/us/en/documents/best-practices\\/cloud-computing-cost-saving-with-a-hybrid-model-brief.pdf}

\bibitem[32]{schellhorn}
Henri Schellhorn :
\textit{A Double Sided Multiunit Combinatorial Auction for
Substitutes: Theory and Algorithms.
}
\texttt{www.researchgate.net/publication/222406473\_A\_double-sided\_multiunit\\\_combinatorial\_auction\_for\_substitutes\_Theory\_and\_algorithms}

\bibitem[33]{clutch}
Clutch:
\textit{Best Cloud Service Providers - 2016 Reviews
}
\\\texttt{clutch.co/cloud}

\bibitem[34]{forbes2}
Clutch:
\textit{The Most Profitable Industries In 2015}
\\\texttt{http://www.forbes.com/sites/liyanchen/2015/09/23/the-most-profitable-\\industries-in-2015/#79ba95517712}


\bibitem[35]{azureprice}
Azure:
\textit{Azure Pricing}
\\\texttt{https://azure.microsoft.com/en-us/pricing/}

\bibitem[36]{awsprice}
Amazon:
\textit{AWS Cloud Pricing}
\\\texttt{https://aws.amazon.com/pricing/\?nc2=h\_ql\_ny\_livestream\_blu}


\bibitem[37]{payghistory}
Arthur M. Anderson:
\textit{The History of the pay-as-you-go policy}
\\\texttt{https://www.jstor.org/stable/1171883?seq=1#page\_scan\_tab\_contents}

\bibitem[38]{onedrive}
OneDrive:
\textit{OneDrive Pricing}
\\\texttt{https://onedrive.live.com/about/en-us/plans/}

\bibitem[39]{dropbox}
Dropbox:
\textit{Dropbox pricing}

\bibitem[40]{spotify}
Spotify:
\textit{Spotify pricing}
\\\texttt{https://www.spotify.com/ch-de/premium/?checkout=false}

\bibitem[41]{marketing}
Zuora:
\textit{THE SUBSCRIPTION PRICING}
\\\texttt{https://www.zuora.com/guides/innovative-subscription-pricing-strategies/}

\bibitem[42]{meng}
Marian Mihailescu and Yong Meng Teo:
\textit{Dynamic Resource Pricing on Federated Clouds}
\\\texttt{https://pdfs.semanticscholar.org/c851/b84ad963ece4310c477c9214f4a1f4fe028a.pdf}


\end{thebibliography}
\clearpage
\addtocmark[2]{Author Index} % additional numbered TOC entry
\renewcommand{\indexname}{Author Index}
\printindex
\end{document}