Cloud Computing Costs and Benefits - Springer Link

1 downloads 2160 Views 1MB Size Report
Nov 6, 2010 - cost data of cloud-based applications which can be used for own cost and effort .... costs ex ante it is very hard to decide for a cloud-service based or a classical data ... structures showed that the following aspects drive primarily costs of a cloud ..... Disaster recovery as a cloud service: Economic benefits.
Cloud Computing Costs and Benefits An IT Management Point of View Nane Kratzke

Abstract Although cloud computing is in all mouth today it seems that there exist only little evidence in literature that it is more economical effective than classical data center approaches. Due to a performed qualitative analysis on COBIT, TOGAF and ITIL this paper postulates that cloud-based approaches are likely to provide more benefits than disprofits to IT management. Nevertheless one astonishing issue is the not often stressed ex ante cost intransparency of cloud based approaches which is a major implicit problem for IT investment decisions. This paper presents considerations how to estimate costs of cloud based systems before they enter their operational phase. This is necessary in order to make economical IT investment decisions for or against cloud computing more objective.

Keywords Cloud • Business information system • Cost • Cost estimation • Cost transparency • ITIL • COBIT • TOGAF

1 Introduction Providing IT-Services is a complex management as well as technological problem. There exist a lot of parameters on different management, design as well as operation levels which have significant influence on the overall effort efficiency. Cloud computing is one of the latest developments within the business information systems domain and describes a new delivery model for IT services based on the Internet, and it typically involves the provision of dynamically scalable and often virtualized resources. N. Kratzke () Computer Science and Business Information Systems, L¨ubeck University of Applied Sciences, L¨ubeck, Germany e-mail: [email protected] I. Ivanov et al. (eds.), Cloud Computing and Services Science, Service Science: Research and Innovations in the Service Economy, DOI 10.1007/978-1-4614-2326-3 10, © Springer Science+Business Media New York 2012

185

186

N. Kratzke

Most of the overall effort efficiency is deduced by capacity efficiency in literature which is intensively proclaimed as a key benefit by cloud service providers. The simple fact that only the used capacity of a cloud-based service has to be paid inveigles to postulate the overall effort effectiveness of cloud-based approaches. Almost every analyzed publication repeats this more or less unreflected—even Talukader et al. [13]. This paper does not denial this postulation but advocates a more critical view. The overall cost effectiveness of can should not be reduced to their capacity efficiency.

1.1 Outline Section 2 starts with a brief summary and quintessence of a performed literature review regarding the actual state of research in cloud cost estimation models. In Sect. 3 the overall relevance of cloud-based approaches is analyzed by well known industry best practice management frameworks (COBIT, TOGAF and ITIL). Section 3 shows furthermore that cloud-based approaches are likely to provide more benefits than disprofits to IT management (see also Kratzke [8, 9]). Nevertheless there exist disprofits and issues which have to be solved. One issue is the ex ante cost intransparency of cloud based approaches which is a major problem for IT investment decisions. This contribution presents in Sect. 4 first considerations how to overcome the issue of ex ante cost intransparency of cloud based services in order to make IT investment decisions for cloud based approaches more reliable and trust worthy.

2 The Literature Review Due to page limitations this paper presents only a short summary of the literature review results. This paper refers to Kratzke [8,9] for a more detailed description of the performed literature review. Last but not least it turned out that no substantial cost estimation models could be found in literature. Weinman provided a “mathematical proof” of the inevitability of cloud computing [16]. Nevertheless Weinman only “proofs” that several usage characteristics (e.g. peak load behaviour) of applications fits very well with the cloud service business model from an economic point of view. He does not provide a model to calculate likely costs of a cloud based information system before it enters its operational phase. This is also stated by Truong and Dustdar [14] formulating the wish “for a cost model associated with application models” provided to academical research communities. But literature review revealed some interesting basic approaches. From this paper point of view the domain specific cost calculation approaches (e.g. Hazelhurst [5] and Berrriman et al. [2]), usage characteristic specific approaches (e.g. Mazhelis et al. [11] did this for communications intensive applications) as well as a domain

Cloud Computing Costs and Benefits

187

neutral indicator based methods (like cost effort per web interaction, e.g. Kossman and Kraska [7]) seem the most promising approaches for providing representative cost data of cloud-based applications which can be used for own cost and effort estimations.

3 Impacts to Well Known Industry Best Practices Krcmar [10] depicts in core three Information Management Domains: Overall Management and Governance Functions, Enterprise Wide Information System Design, Information Systems Development1 and Information Systems Operation (see Fig. 1). This paper covers all mentioned IT management domains by three industry best practice standards (COBIT, TOGAF, ITIL). So this section focus the impact of cloud computing to overall governance functions by using COBIT [6], to enterprise wide information systems design by using TOGAF [4] and to information systems operation by using ITIL (according to B¨ottcher [3]) as a evaluation reference. By applying these models qualitative impacts of business cloud computing to one or more of the mentioned industry best practices standards are deduced. For each of the mentioned models a process tree was developed and used to depict qualitative impacts to the mentioned reference models. These process trees are used to depict qualitative impacts. An qualitative impact may be positive (effort reducing), negative (effort adding) or neutral (effort invariant). These impacts are rated in the following way: – Positive (marked (C)) if cloud computing may reduce efforts (compared to classical information system approaches). – Negative (marked ()) if cloud computing introduces additional efforts (compared to classical information system approaches).

Fig. 1 Reflected IT management standards and classification models

1

System design and development is not covered by this paper.

188

N. Kratzke

– Neutral (no mark) if cloud computing has no effect (compared to classical information system approaches).2 Due to page limitations this paper presents no detailed (but in references existing) reasoning of the postulated qualitative impacts to the analyzed models. Only the core impacts with a short reasoning are stated. And whenever a process is rated neutral (and therefore not mentioned in the next sections) this paper states neither effort adding nor reducing impacts to these processes. This is mainly due to the fact that these process steps should be done with or without a cloud-based foundation of (business) information systems.

3.1 Impact to COBIT (Governance) The Control Objectives for Information and related Technology (COBIT) is a set of best practices (framework) for information technology (IT) management. COBIT provides a set of measures, indicators, processes and best practices, to assist maximizing the benefits derived through the use of information technology, and developing appropriate IT governance and control in a company. COBIT defines a set of deliver and support, acquire and implement, monitor and evaluate as well as planning processes to operationalize IT-governance in companies (see Fig. 2).

3.1.1 Cloud Computing Is Likely to Reduce Overall Efforts In the Following COBIT Process Steps Within the deliver and support process it is likely to reduce efforts for managing performance and capacity, operations, continuous service as well as managing physical environment due to the fact that this tasks are transferred to the cloud vendor (Talukader et al. [13]). Furthermore efforts are likely reduced in definition of third party services (already done by cloud vendor) and identification and allocation cost (also done by cloud vendor during billing—so called ex post cost transparency). Regarding the acquire and implementation process it is likely to reduce efforts in identifying automated solutions, acquire and maintain technology infrastructure and procure IT resources due to the fact that these tasks have to be performed by the cloud vendor [13]. Regarding the planning processes it is likely to reduce efforts in managing IT investments (this has to be done by the cloud vendor) as well as IT Human resources (tendency to reduce the IT staff but need for IT professional with 2

This is mainly due to tasks which are necessary for cloud-based or classical business information systems governance, design, development or operation as well.

Cloud Computing Costs and Benefits

189

Fig. 2 Qualitative cloud impact to the Cobit process tree

cloud knowledge). Regarding the monitor and evaluation processes it is likely to reduce efforts in monitoring and evaluating IT performance (this has to be done by the cloud vendor).

3.1.2 Cloud Computing Has the Tendency to Enhance Occasionally Efforts In the Following COBIT Process Steps Within the deliver and support process it is likely to create additional efforts due to a lot of security issues, (see Onwubiko et al. [12]) as well as due to a more complex configuration management of virtual cloud assets which are not under direct control of the cloud customer. Regarding the acquire and implementation process it is likely to create additional efforts due to more complex (PaaS based) Application development and their corresponding installation and accreditation processes. Regarding the planning processes it is likely to create additional efforts

190

N. Kratzke

in assessing IT risks, defining IT processes and relationships as well as managing projects (due to an additional actor—the cloud vendor). Regarding the monitor and evaluation processes it is likely to create additional efforts ensuring compliance with external requirements (due to the fact that the cloud vendor and its internal processes have to reflected, see Onwubiko et al. [12]).

3.2 Impact to TOGAF (Enterprise Wide Design) The Open Group Architecture Framework (TOGAF) is a framework for enterprise architecture management which provides a comprehensive approach to the design, planning, implementation, and governance of an enterprise information architecture. TOGAF based Enterprise Architectures are typically modeled at four levels: Business, Application, Data and Technology. Application and data architecture are integrated in a so called Information System Architecture. TOGAF Enterprise Architectures should be developed using the Architecture Development Model (ADM) Cycle shown in Fig. 3. TOGAF relies deeply on modularization, standardization and already existing, proven technologies and products which fits well with standardized cloud based IaaS, PaaS or SaaS services. By using cloud-based approaches it is likely to reduce application design efforts by using SaaS or PaaS3 cloud-based services4 due to the fact that the cloud service providers have to provide precisely defined architecture building blocks which are there chargeable assets. It is furthermore likely to reduce technology architecture design efforts due to the fact that they are predefined by IaaS5 cloud service providers. In the most of use cases it is easier to chose a technology architecture than to design one. Both above mentioned facts will likely produce new opportunities and solutions for business information systems and their corresponding information architectures—this is also mentioned by Barr who emphasizes the business characteristics of cloud computing: business flexibility, cost associativity as well as cheap experimentation (see Barr [1], p. 9 and 10). According to the performed literature research no negative effects of cloud computing could be identified from a TOGAF point of view.

3

SaaS—Software as a Service, e.g. SAP BUSINESS BYDESIGN; PaaS—Platform as a Service, e.g. Google Apps. 4 See [12–14] for a definition of SaaS or PaaS. 5 IaaS—Infrastructure as a Service, e.g. Amazon EC2 (see [12–14] for a definition of IaaS).

Cloud Computing Costs and Benefits

191

Fig. 3 Qualitative cloud impact to the TOGAF process map

3.3 Impact to ITIL (Operations) ITIL is an industry best practice standard for operating and providing IT-services according to B¨ottcher [3]. ITIL provides best practice processes to design and operate IT-services for internal or external customers. IT-Services are driven by general business requirements supporting a service strategy. All IT services are handled in a service pipeline defining planned, operated and outdated processes. The step from planned to operated services is done by so called service transition processes. Figure 4 lists all relevant ITIL processes according to B¨ottcher [3].

192

N. Kratzke

Fig. 4 Qualitative cloud impact to the ITIL V3 process tree

3.3.1 Cloud Computing Shows the Potential to Reduce Overall Efforts In the Following ITIL Process Steps By using cloud-based approaches it is likely to reduce service design efforts in capacity, availability as well as continuity management (see e.g Wood et al. [17] or Talukader et al. [13]). This is due to the inherent capabilities of clouds. It is furthermore likely to reduce service operation efforts in event, incident as well as problem management because a lot of efforts have to be handled by cloud service providers.

Cloud Computing Costs and Benefits Table 1 Overall weaknesses and strengths of cloud based approaches Derived by analyzing COBIT Strengths Inherent scalability in capacity and performance x Inherent continuousity and availability Ex post cost transparency x Provision of automated infrastructure services x Provision of automated functional services x Physical infrastructure free (for customers) x Low level service free (for customers) Higher order service enabling x Weaknesses Additional cloud SW development skills x More complex configuration management x More complex service and process management x More complex security management x More complex compliance management x Ex ante cost intransparency x

193

TOGAF

x x

ITIL x x x x x x x

x x x x

3.3.2 Cloud Computing May Occasionally Enhance Efforts in the Following ITIL Process Steps By using cloud-based approaches it is likely to increase service level management efforts which is due to involving an additional service providing party (the cloud service provider, check Talukader et al. [13]). Additional efforts are also likely to perform information security and compliancy management (see Onwubiko et al. [12]) because aspects like privacy, data ownership, confidentiality, data location, regulatory compliance, forensic evidence, auditing and overall trust issues have to be considered. Furthermore additional efforts are likely to perform a service asset and configuration management. A configuration management has to handle and control virtual cloud assets which are not under direct control of the cloud-using service customer.

3.4 Qualitative Weaknesses and Strengths of Clouds Regarding the cloud impacts to COBIT (Governance, see Sect. 3.1), to TOGAF (Enterprise Wide Systems Design, see Sect. 3.2) and to ITIL (Service Operation, see Sect. 3.3) this paper postulates the in Table 1 listed overall qualitative weaknesses and strengths of cloud-based approaches to IT management. Analyzing Table 1 you see that the strengths of clouds lay in their inherent structure (scalability, continuousity, availability, etc.) as well as necessary market requirements (provide well defined and therefore billable infrastructure or functional services) which reduce efforts on the cloud customer side (avoiding to provide such services on their own with smaller economical scale effects).

194

N. Kratzke

The weaknesses according to Table 1 are mainly introduced by the fact that an additional player (the cloud service provider) enters the game—so additional interaction business processes become necessary introducing additional efforts. From this paper point of view these additional service, process and configuration management efforts will be overcompensated by the strengths of the cloud based approaches. Sections 3.1–3.3 showed that more processes have benefits than disprofits. But let us look closer to security and compliance management aspects. This category of weakness may come along with substantial “showstoppers” for a cloud based approach. Whenever a company has to be compliant to regulatories which can not be fullfilled by cloud service level agreements (e.g. privacy requirements, data ownership, confidentiality, data location, forensic evidence, auditing, etc.) cloudbased approaches may be not feasible. But this is not due to economical but higher order considerations. Nevertheless there exist even an ex ante cost transparency weakness as it is stated for example by Truong and Dustar [14]. This very important weakness (from an IT management point of view) is even little reflected in literature so far. To answer the question whether a cloud-based approach is more cost efficient than a classical data center centric approach it has to be answered the question what costs will be generated per month before an application enters operation (see also Walker et al. [15]). This is very difficult to answer ex ante because it is influenced by a bunch of interdependent parameters. Some of them are analysed in the following Sect. 4. This finding is astonishing because it is postulated and repeated by several authors that cloud services are increasing cost transparency (e.g. Talukader et al. [13]). This paper agrees that cloud services will increase ex post cost transparency mainly due to the underlying billing process of cloud service providers. But it seems very hard to estimate cloud costs ex ante. Nevertheless this is needed for IT investment decisions. Without being able to calculate or estimate cloud service costs ex ante it is very hard to decide for a cloud-service based or a classical data center centric approach.

4 A Resulting Cost Estimation Approach As it was stated in Sect. 3.4 cloud services provide excellent ex post cost transparency. It was furthermore stated that there exist barley cost estimation models6 for ex ante cost calculation and transparency. This paper presents an approach to use the strength of ex post cost transparency of cloud services to compensate the weak of ex ante cost estimation.

6

Which are cloud vendor independent.

Cloud Computing Costs and Benefits

195

b

a

System space

Using ex post cost data to estimate costs

Fig. 5 Visualization of the cost estimation principle

The core idea is a very simple one. Whenever running a cloud-service based system it is easy to gather the costs ex post. Your cloud service provider will deliver a bill with the used cloud service assets. Whenever you plan another cloud service based system of comparable complexity and usage parameters you can look at costs of your already running cloud service based system. It is likely that your ex post costs of the existing system will have the same characteristics of your planned system of comparable complexity and usage parameters. This very simple idea has one evident problem. It will provide only good cost estimations for comparable systems with comparable usage characteristics which is not a very realistic assumption. But what to do when decisions have to be made for non comparable systems? We have to make our model a little bit more complex. One possibility is to inter- or extrapolate costs from nearest neighbors (see Fig. 5). Nevertheless we have to describe cost driving parameters in a way that they can be inter- or extrapolated to your planned system and we have to deduce parameters which are appropriate to describe the dimensions of a system space (which are most likely much more than two—so Fig. 5 shows an extreme simplification of the to be encountered problem). A substantial cost calculation model should have the capability to select the most comparable system of a given system space in order to inter- or extrapolate the most appropriate cost driving parameters for a cloud based application.

4.1 A Cloud Cost Model A performed analysis of cloud service providers and their underlying billing structures showed that the following aspects drive primarily costs of a cloud based approach (see Fig. 6).

196

N. Kratzke

Fig. 6 Influences to cloud based costs

Figure 6 shows the principal relations which should be covered by cloud based cost analysis. First of all a perfect scalable (and cloud based) information system should produce no costs at all if the system do not process any requests. And the costs should raise with the amount of to be processed requests. So the core cost driver are system usage requests. System usage requests create a cloud service usage on an IaaS, PaaS or SaaS level (we use the short term XaaS if we refer to all three levels) which is lastly billed by cloud service providers. Nevertheless which types of XaaS services are used (and therefore billed) are also highly influenced by the general (cloud based) information systems architecture. Furthermore there exist feedback relations in the presented model. System architecture also influences the generated service usage. Imagine a non scalable system which has to handle a spur-of-the-moment usage peak. Typically the system response times increase dramatically (or fail completely) if the peak usage exceed significantly the designed maximum capacity. If this happens to often it is very likely that the overall system usage declines in total because the system is rated as unreliable by its users. Such spur-of-the-moment peak loads show no significant impact to the overall system usage of highly scalable information systems because theses systems can handle peak load scenarios. So you can see an indirect feedback relation between a system architecture and the overall system usage of a cloud based information system.

Cloud Computing Costs and Benefits

197

Another indirect feedback relation exists between the generated costs and the system architecture. If a cloud based information system is meant to be to expensive it is very likely to change its cost characteristic by changing its system architecture—typically to reduce costs. There exist several strategies to do this— e.g. replacing typical more expensive PaaS services by less convenient but cheaper IaaS services, using more load balanced but less performant processing instances, etc. This all has impact to the overall cloud based information system architecture. In the following sections it is shown what parameters are appropriate to measure cloud service usage and corresponding costs (see Sect. 4.2) and how influence of an information systems architecture can be expressed in numbers (see Sect. 4.4) as well as how to measure the overall system usage of a cloud based information system (see Sect. 4.3). Finally Sect. 4.5 shows how these components can be used to identify comparable cloud based systems in order to use their ex post cost and usage data to estimate costs of a planned cloud based information system.

4.2 Parameters to Describe Service Usage and Resulting Costs Analysis of real world bills of cloud service providers7 showed that on almost each cloud service provision level (IaaS, PaaS, SaaS) cloud customers are billed for the following service usage categories: Data Transfer This includes all data which has to be transferred into, out of or within a service. Sometimes data transfer is made explicit by being billed for incoming, outgoing and inner traffic. Data Storage This includes all data which has to be stored by a service provider in order to process service requests. Processing This includes the amount of time processing instances were used. Processing time can be billed by instance uptime hours (typically only on IaaS level) or by time spent for processing requests (this is more common on PaaS and SaaS level). Requests If a service provides its output via requests it is a common strategy to be billed per request. This is very common on PaaS and SaaS level but less common on IaaS level. Nevertheless even on IaaS level this type of billing can be found— especially in data backup services like EC2 from Amazon Web Services. This is called a micro request. Micro requests are seldom relevant cost drivers.

7 The billing of Amazon Web Services and Google App Engine was analysed intensively but the derived findings stay valid for other cloud service providers like Rackspace.com, Salesforce.com, Windows (Azure), etc. cross checking their public accessible billing customer informations.

198

N. Kratzke

Table 2 Cost categories aligned to service levels

Cost category

IaaS

PaaS

SaaS

Data transfer (in, out, within) Data storage Processing (instance hours) Processing per request Micro request Request Network

x x x

x x

x x

x

x

x

x

x x

Network This includes all efforts in order to get a necessary network infrastructure running. Typically network costs are billed only on the IaaS level. If they are relevant on higher service levels they are typically billed via request costs.8 Typically you are billed for things like load balancers, ip addresses, auto scalers, etc. The following Table 2 shows which cost categories apply typically on which cloud service level. By applying the following table it is possible to compare different cloud service providers. It is now possible to aggregate costs of different cloud service providers along a row or a column. Let us do this exemplarily by aggregating Platform as a Service costs of a specific cloud service provider. As you can see in Table 2 Platform as a Service costs are composed of data transfer, data storage, processing per request and request costs. Typically these parameters are billed by cloud services providers per identifiable service (e.g. AWS Messaging Service). – – – –

Amount of service requests SERV REQSt to a service t Data storage SERV STORAGEt associated with service t Data transfer SERV TRANSFERr associated with service t Resulting computing hours PROCHOURSt associated with service t

These parameters can be summarized over all used services t of a cloud-based system and multiplied with the charge per service usage parameter CSERV REQ , CSERV STORAGEt , CSERV TRANSFERt and CPROCHOURt . Therefore these costs can be simply aggregated to system service cost PAASCOSTt per month for a service t. PAASCOSTt D

X

! SERV REQSti

CSERV REQt C

i

C

X i

! SERV TRANSFERti

X

! SERV STORAGEti

CSERV STORAGEt

i

CSERV TRANSFERt C

X

! PROCHOURSti

CPROCHOURt

i

For example this is what Amazon Web Services is charging on a monthly basis for its S3, RDS-Service or Google for its GoogleApp-Service.

8

Typically in these cases network costs are billed as non mentioned cost component of request costs in bills.

Cloud Computing Costs and Benefits

199

The same is possible for infrastructure IAASCOST or software as a service costs SAASCOST .9 So it is possible to compare the costs of two or more different cloud service providers on different service levels. It is furthermore possible to calculate service level independent costs for network NETWORKCOST , request REQUESTCOST , processing PROCESSINGCOST , data transfer DATA TRANSCOST or data storage DATA STORAGECOST costs if you are interested in separating these concerns.

4.3 Parameters to Describe System Usage In Sect. 4.4 it is shown by principle how to compare different architectures in order to get a feeling whether a reference cloud-based system has a comparable architecture and therefore is a good candidate to evaluate its ex post cost data. But systems with similar architectures do not automatically produce the same cost. Cloud costs are produced primarily by usage.10 We only use architectures to select the most comparable systems due to the fact that we think they will produce a similar cost characteristic. The following usage parameters are deduced by the analysis of web-based systems and analysis tools like GoogleAnalytics,11 AWStats,12 or Open Web Analytics.13 So the following considerations are strictly speaking only valid for web based systems but the core idea should be easily transferred to other types of systems. All mentioned analytic packages provide a usage analysis of web-based systems and they all distinct (in common) the following levels of a user interaction: – Number of page views in a given time frame (typical within a hour, day, month, etc.) as smallest entity – Number of visits in a given time frame (a visit aggregates all page views produced by the same visitor in a given time frame) – Number of users (user of a web-based system with an account) The cost producing usage is only induced in a web-system by pageviews. Nothing else. From a cost perspective it is irrelevant whether 1,000 pageviews are produced by 100 or 10 visits which can be assigned to 100, 10 or only 1 visitor (user). The produced costs are the same. Nevertheless from cost estimation perspective it is much easier to estimate users or visits than to estimate page views.14

9

Which is not done in this paper due to page limitations. But the aggregation is analog. This is what every cloud service provider stresses as THE key benefit of cloud computing. 11 http://www.google.com/intl/en/analytics/index.html. 12 http://awstats.sourceforge.net. 13 http://www.openwebanalytics.com. 14 Think about a situation like, setting up a social network and you plan to orient on a platform like facebook.com or linkedin. Facebook has something about five hundred million users, linked in has 10

200

N. Kratzke

We think that the above mentioned web-based usage parameters can be generalized and that they are adequate for describing and compare usage characteristics. They are all measured per month, due to the fact that underlying billing cycles of cloud service providers are also made on a monthly interval. – Number of requests reqs per month (generalized from page views) to a system – Number of user interactions ints per month (generalized from visits) with a system – Number of active accounts accs in a month (generalized from users) of a system – Number of unregistered users anons in a month (generalized from users) of system For cost estimation and inter- or extrapolation total numbers should be made relative. So typically interactions per user ipu and cost driving requests per interaction rpi are calculated. Especially for web-based systems these data can be easily measured by using established toolings already mentioned above. ipu WD

anons C accs ints

rpi WD

reqs ints

By having these indicators we can now map the core cost driver (requests) to the usage data which is measured and billed by the cloud service provider. This is done in Sect. 4.5.

4.4 Parameters to Describe Architectural Influence The architecture of a cloud-based system influences massively the costs. For example a system running on a single instance is likely to produce much less costs then a massive parallel systems with a lot of instances. Nevertheless between these extremes there exist hybrid forms and it is not always obvious which configuration is more cost efficient.15 For first considerations a system can be described by a set of characteristic parameters—for example the following ones: – – – –

Total number of instances I Number of additional (autoscaling) instances IAS Number of running loadbalancers L Set of used services S

about seventy million users. But how many pageviews has Facebook or linkedin? Hard to estimate or to research. 15 For example answer yourself the question whether two high-performance instance systems are more cost efficient than ten low-performance but load balanced ones? It is likely that your answer is like: “It depends.”

Cloud Computing Costs and Benefits

201

So the above mentioned architectural description tuple seem adequate to compare different cloud-based system architectures on a basic infrastructure level by the following tuple .I; IAS ; L; S /. It is possible to define a similarity function to compare two architecture description tuples .I1 ; IAS1 ; L1 ; S1 / and .I2 ; IAS2 ; L2 ; S2 /. Two architectures are likely to produce the same cost characteristics if the have the same amount of running instances load balancers and are using the same services. This is expressible by the following function definition16 : similarity.I1 ; IAS1 ; L1 ; S1 ; I2 ; IAS2 ; L2 ; S2 /   IAS1 C 1 L1 C 1 jS1 \ S2 j 1 I1 C 1 C C C WD 4 I2 C 1 IAS2 C 1 L2 C 1 max.jS1 [ S2 j; 1/ So two identical architectures will produce a similarity of 1. The more different architectures are the more their similarity will move away from 1. Such similarity functions can be used to filter most comparable systems with ex post data of a given cost database (please compare Fig. 5).

4.5 Using Cloud Service and System Usage As Well As Architectural Description Parameters to Estimate Costs So architectural parameters are used to filter most comparable systems out of a given system space by using a similarity function (see Sect. 4.4 for an example). We furthermore have identified the core cost driving parameter to a cloud-based system—the request. So requests have to be related to the usage data and resulting cloud costs. As it was mentioned in Sect. 4.2 the total costs can be calculated by summing up service level costs17 or category costs: TOTAL D NETWORKCOST C REQUESTCOST C PROCESSINGCOST C DATA TRANSCOST C DATA STORAGECOST Reflecting all made considerations it is now possible to calculate total cost estimations for a planned or estimated number of users per month. The necessary cost indicators are collected but measuring costs and usage characteristics of already running cloud-based systems which are comparable in architecture (see Sect. 4.4). Please have the definitions of rpi and ipu in Sect. 4.3 in mind.

16

This is due to validation—it might be possible to use a completely different similarity function. The presented function is only for exemplification purposes. 17 TOTAL D IAASCOST C PAASCOST C SAASCOST .

202

N. Kratzke

TOTAL  rpi  ipu  user r PROCESSINGCOST PROCESSINGestimate.user/ D  rpi  ipu  user r DATA STORAGECOST STORAGEESTIMATE .user/ D  rpi  ipu  user r TOTALestimate .user/ D

At the L¨ubeck University of Applied Sciences we use the presented approach successfully in order to estimate cloud costs for the next semester.18

5 Conclusions, Outlook and Acknowledgements For IT management investment decisions an ex ante rather than an ex post cost transparency is needed. But ex ante cost estimation models do not exist so far and have to be established and cross checked. We presented a cost estimation model in its early research stages by using the strengths of cloud services (ex post cost transparency) to provide missing ex ante cost transparency in order to improve economical IT management decision-making for or against cloud based information system solutions. The presented approach is in its core idea quite simple but powerful. – Measure service usage data of already running systems (costs for infrastructure, scalability efforts and additional service usage). – Measure usage characteristics (requests, interactions and user amounts). – Analyse the architecture of a planned system and already running systems in the cloud in order to find out the most comparable system by a similarity function. – Deduce the most comparable real world systems to estimate the cloud cost of planned systems by using their usage—as well as cost-indicators. Thanks to Amazon Web Services for supporting our ongoing research in this field with several substantial research as well as educational grants.

18

The cost estimation is performed for lectures “Web Technology” and “Databases”. In both lectures a varying amount of students over time have to pass practical exams in which they have to set up and implement an interactive web presence or database intensive application. The system usage characteristic includes a 24  7 phase of 4 weeks. All necessary infrastructure is provided by Amazon Web Services. By applying the mentioned cost estimation models we figured out to have 18.81 USD per student cloud costs per student. So if our next classes have 100 students we assume 1,881.00 USD for 100 students cloud costs for the next semester.

Cloud Computing Costs and Benefits

203

References 1. Barr, J.: Host your Web Site in the Cloud. sitepoint (2010) 2. Berriman, G. B., Juve, G., Deelman, E., Regelson, M.: The application of cloud computing to astronomy. In: Proceedings of the e-Science in Astronomy Conference (2010) 3. B¨ottcher, R.: IT-Servicemanagement (ITIL V3). Heise Verlag, Hannover (2008) 4. Group, O.: The Open Group Architecture Framework (TOGAF). Van Haren Publishing, 8th edn. (2007) 5. Hazelhurst, S.: Scientific computing using virtual high-performance computing. In: Proceedings of the 2008 annual research conference SAICIST’08 (2008) 6. ISACA: Cobit 4.1. IT Governance Institute (2008) 7. Kossmann, D., Kraska, T.: Data management in the cloud. Datenbank Spektrum published online 06.11.2010 (2010) 8. Kratzke, N.: Cloud-based it management impacts. In: Proceedings of the 1st Internation Conference on Cloud Computing and Services Science (CLOSER 2011). pp. 145–151 (2011) 9. Kratzke, N.: Overcoming ex ante cost intransparency of clouds. In: Proc. of the 1st international Conference on Cloud Computing and Services Science (CLOSER 2011, special session on Business Systems and Aligned IT Services - BITS 2011). pp. 707–716 (2011) 10. Krcmar, H.: Informationsmanagement. Springer, 5th edn. (2010) 11. Mazhelis, O., Tyrv¨ainen, P., Eeik, T. K., Hiltunen, J.: Dedicated vs.. on-demand infrastructure costs in communications-intensive applications. In: Proc. of the 1st international Conference on Cloud Computing and Services Science (CLOSER 2011), pp. 362–370 (2011) 12. Onwubiko, C.: Cloud Computing: Principles, Systems and Applications, chap. 16 Security Issues to Cloud Computing, pp. 271–288. N. Anatonopoulos and L. Gillam (2010) 13. Talukader, A. K., Zimmermann, L., Prahalad, H.: Cloud economics: Principles, costs and benefits. In: Cloud Computing - Computer Communications and Networks. vol. 4, pp. 343–360. Springer (2010) 14. Truong, H.-L., Dustdar, S.: Cloud computing for small research groups in computational science and engineering. Computing. vol. 91(1), 75–91. Springer (2011) 15. Walker, E., Brisken, W., Romney, J.: To lease or not to lease from storage clouds. Computer pp. 44–50 (April 2010) 16. Weinmann, J.: Mathematical proof of the inevitability of cloud computing 17. Wood, T., Cecchet, E., Ramakrishnan, K. K., Shenoy, P., van der Merwe, J., Venkataramani, A.: Disaster recovery as a cloud service: Economic benefits. In: HotCloud’10 Proceedings (2010)