A game-theoretic approach to the financial benefits of infrastructure-as-a-service
Introduction
When a client considers cloud computing, a variety of factors like privacy concerns and strategic decisions have to be reckoned with the particular case. In particular, costs are a key factor in the decision process. Case studies that ask whether or not processing in the cloud is feasible usually only regard current prices for cloud services. Since the cloud market develops, assuming constant prices is insufficient for long-term decisions. This paper abstracts from current market prices and investigates the interaction of cloud provider and clients from an analytical perspective. A general understanding of how providers and clients potentially benefit financially from Infrastructure-as-a-Service (IaaS) can help clients to appraise price uncertainty in strategic resource planning decisions. Providers gain insights on how pricing and charging models affect service usage.
Our analysis focuses especially on the combined use of cloud services and an own data center, which offers a variety of possibilities how clients may split up their processing demand. While cloud service prices are most likely considered in the resource allocation decision of a client, it is unknown how this interrelation affects future cloud pricing.
Market dynamics depend on provider and client behavior. By contrasting the possible actions of these market actors, game theory can identify stable market situations that suggest likely or advisable behavior. We contribute a game-theoretic model and determine its equilibria in order to estimate future pricing and expected usage of IaaS in hybrid cloud scenarios. We further discuss the impact of factors like load distribution and economies of scale on the model. Also, the effects of a simultaneous offer of reserved instances is explored.
This paper is organized as follows. The contribution of this paper is put into relation to other research in Section 2. A market model for on-demand cloud infrastructure is developed throughout Section 3. In Section 4, the model is applied to an example case. Section 5 discusses the impact of a reserved charging option on the market. A general discussion of research results is presented in Section 6, which also concludes the paper.
Section snippets
Related work
Several publications deal with the suitability of cloud services as a substitute for on-site corporate IT. Guidelines for the decision process like [1], [2] usually include a financial comparison of feasible solutions. Calculation models for total cost of ownership (TCO) of a data center [3], [4] can be taken as a basis for this. There are also ready-to-use calculators [5] for a direct comparison of expected costs based on specified demand. A cost model specifically for hybrid clouds is
Setup
A game-theoretic model of an on-demand infrastructure cloud market is suggested in the following; the goal is to estimate future pricing. The model is set up as one player being a provider that offers an on-demand computation instance. Such an instance provides capacity for processing and storage like a physical server and fees apply only when the instance is in use. The other player is a client that may utilizes the offered product. An extensive form game [17] is used since the provider
Case study
The German IT magazine iX published a case study for a hypothetical company in which the TCO of a new data center was compared to the costs of a co-location setup and the use of Amazon’s EC2 service [3]. In the study, costs per year for an owned data center consist of investment cost amortization and running costs. Investments are acquisition costs for server and network hardware and operation system licenses (3 years write-off) as well as infrastructure and building costs (15 years
On-demand and reserved instances
The presented model considers pricing and usage of on-demand instances, which means that they only have to be paid for the time in use. They can be a cost-saving substitute for a client’s data center because the number of instances in use can be adapted to current demand. Due to statistical multiplexing, the hardware necessary to meet demand of all clients is fewer in a cloud environment than when the clients run their own data centers. This is the main benefit of the cloud next to economies of
Discussion & conclusions
This section generally discusses the presented model and briefly addresses a few questions that were raised earlier in the paper. It also presents starting points for further research.
Acknowledgments
This work is partially supported by the German Research Foundation (DFG) within the Research Training Group Automatisms (GRK 1479) and the Collaborative Research Center On-The-Fly Computing (SFB 901).
Jörn Künsemöller: 10/2004–09/2007/10/2007–03/2010: Student in the bachelor/master program Informatics in the Natural Sciences (NWI) at the University of Bielefeld.
06/2010–05/2013: Member of the Research Training Group Automatisms at the University of Paderborn.
06/2013–11/2013: Research assistant in the Collaborative Research Center On-The-Fly-Computing, University of Paderborn.
Since 02/2014: Software Engineer at the radio astronomy group of the University of Bielefeld.
References (25)
- et al.
An experimental analysis of ultimatum bargaining
J. Eco. Behav. Organ.
(1982) - et al.
Do clouds compute? A framework for estimating the value of cloud computing
- S. Lamberth, A. Weisbecker, Wirtschaftlichkeitsbetrachtungen beim Einsatz von Cloud Computing, in: Vom Projekt zum...
- C. Christmann, J. Falkner, D. Kopperger, A. Weisbecker, Schein oder Sein, iX special: Cloud, Grid, Virtualisierung,...
- J. Koomey, A simple model for determining true total cost of ownership for data centers (White Paper), Uptime...
- Amazon Web Services, Amazon EC2 Cost Comparison Calculator, Website. URL...
- M.M. Kashef, J. Altmann, A cost model for hybrid clouds, in: Proc. 8th international conference on Economics of Grids,...
- et al.
Collocation Games and Their Application to Distributed Resource Management, Tech. Rep.
(2009) - et al.
A game-theoretic method of fair resource allocation for cloud computing services
J. Supercomput.
(2010) - P. Dube, R. Jain, C. Touati, An analysis of pricing competition for queued services with multiple providers, in: ITA...
Cited by (10)
Economics of Computing Services: A literature survey about technologies for an economy of fungible cloud services
2018, Future Generation Computer SystemsEconomics of computing services
2016, Future Generation Computer SystemsEconomics of computing services
2014, Future Generation Computer SystemsEfficient Resource Scheduling for Distributed Infrastructures Using Negotiation Capabilities
2023, IEEE International Conference on Cloud Computing, CLOUDTwo-stage pricing strategy for personal cloud storage: free trial and the cloud security risk
2022, International Journal of Sensor Networks
Jörn Künsemöller: 10/2004–09/2007/10/2007–03/2010: Student in the bachelor/master program Informatics in the Natural Sciences (NWI) at the University of Bielefeld.
06/2010–05/2013: Member of the Research Training Group Automatisms at the University of Paderborn.
06/2013–11/2013: Research assistant in the Collaborative Research Center On-The-Fly-Computing, University of Paderborn.
Since 02/2014: Software Engineer at the radio astronomy group of the University of Bielefeld.
Holger Karl: 1996: Master in computer science from the Technical University Karlsruhe (advisor Prof. Ungerer, now at University Augsburg).
1999: Ph.D. in computer science from the Humboldt University of Berlin (advisor Prof. Malek).
Studied abroad at University of Massachusetts at Amherst (1993/1994) and New York University (1997/1998).
2000–2004: Research assistant (“Wissenschaftlicher Assistent”) at Telecommunication Networks group (Prof. Adam Wolisz) of Technical University Berlin.
Since 10/2004: Full professor of practical computer science/computer networks at University of Paderborn, Department of computer science.