Abstract
In Amazon EC2, cloud resources are sold through a combination of an on-demand market, in which customers buy resources at a fixed price, and a spot market, in which customers bid for an uncertain supply of excess resources. Standard market environments suggest that an optimal design uses just one type of market. We show the prevalence of a dual market system can be explained by heterogeneous risk attitudes of customers. In our stylized model, we consider unit demand risk-averse bidders. We show the model admits a unique equilibrium, with higher revenue and higher welfare than using only spot markets. Furthermore, as risk aversion increases, the usage of the on-demand market increases. We conclude that risk attitudes are an important factor in cloud resource allocation and should be incorporated into models of cloud markets.
Part of this work was completed while D. Hoy was an intern at Microsoft Research.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
Notes
- 1.
This might more naturally be called a “reservation market” and we switch to this terminology in the remainder of the paper; however we stick to the term “on-demand” for the current discussion as this is the term used by Amazon.
- 2.
Of course, this model abstracts away from many reasonable sources of risk aversion in the cloud, such as clients with diminishing marginal returns for multiple instances, the cost of prematurely terminating a long-running task. Even ignoring these factors, our model still generates heterogeneous preferences toward on-demand versus spot pricing.
- 3.
Since our model abstracts away from inter-temporal effects, we do not explicitly model the impact of fluctuating spot prices and changes to on-demand prices over time. Investigating a repeated-game model of the market, and/or agents with time-dependent preferences (e.g., minimizing the cost of a large job subject to a deadline), is left as a direction for future work.
- 4.
We define risk aversion with respect to agent preferences directly, rather than via the Arrow-Pratt measure, to avoid requiring utility curves be twice differentiable.
- 5.
This price may not be unique if \(q = 0\) or \(q=1\). In these cases we define \(p_{s}(q)\) to be the supremum of prices satisfying the written condition, which will be \(\infty \) for \(q = 0\).
References
Abhishek, V., Kash, I.A., Key, P.: Fixed and market pricing for cloud services. CoRR, abs/1201.5621(2012)
An, B., Lesser, V., Irwin, D., Zink, M.: Automated negotiation with decommitment for dynamic resource allocation in cloud computing. In: Proceedings of the 9th International Conference on Autonomous Agents and Multiagent Systems, vol. 1, pp. 981–988. International Foundation for Autonomous Agents and Multiagent Systems (2011)
Ausubel, L.M., Cramton, P.: Using forward markets to improve electricity market design. Util. Policy 18(4), 195–200 (2010)
Ben-Yehuda, O.A., Ben-Yehuda, M., Schuster, A., Tsafrir, D.: Deconstructing amazon EC2 spot instance pricing. ACM TEAC 1(3), 1–20 (2013)
Borgs, C., Candogan, O., Chayes, J., Lobel, I., Nazerzadeh, H.: Optimal multiperiod pricing with service guarantees. Manag. Sci. 60, 1792–1811 (2014)
Fu, H., Hartline, J., Hoy, D.: Prior-independent auctions for risk-averse agents. In: EC 2013(2013)
Hu, A., Matthews, S.A., Zou, L.: Risk aversion and optimal reserve prices in first- and second-price auctions. J. Econ. Theory 145(3), 1188–1202 (2010)
Ma, D., Huang, J.: The pricing model of cloud computing services. In: EC 2012, pp. 263–269 (2012)
Maskin, E., Riley, J.: Optimal auctions with risk averse buyers. Econometrica 52(6), 1473–1518 (1984)
Mathews, T., Katzman, B.: The role of varying risk attitudes in an auction with a buyout option. Econ. Theory 27(3), 597–613 (2006)
Matthews, S.A.: Selling to risk averse buyers with unobservable tastes. J. Econ. Theory 30(2), 370–400 (1983)
Matthews, S.A.: Comparing auctions for risk averse buyers: a buyer’s point of view. Econometrica 55(3), 633–646 (1987)
Menache, I., Shamir, O., Jain, N.: Dynamic resource allocation for executing batch jobs in the cloud. In: ICAC 2014
Pratt, J.W.: Risk aversion in the small and in the large. Econometrica 32(1), 122–136 (1964)
Riley, J., Samuelson, W.: Optimal auctions. Am. Econ. Rev. 71(3), 381–392 (1981)
Wang, W., Li, B., Liang, B.: Towards optimal capacity segmentation with hybrid cloud pricing. In: ICDCS 2012, pp. 425–434. IEEE (2012)
Zhang, H., Li, B., Jiang, H., Liu, F., Vasilakos, A.V., Liu, J.: A framework for truthful online auctions in cloud computing with heterogeneous user demands. In: IEEE INFOCOM, pp. 1510–1518. IEEE (2013)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2016 Springer-Verlag GmbH Germany
About this paper
Cite this paper
Hoy, D., Immorlica, N., Lucier, B. (2016). On-Demand or Spot? Selling the Cloud to Risk-Averse Customers. In: Cai, Y., Vetta, A. (eds) Web and Internet Economics. WINE 2016. Lecture Notes in Computer Science(), vol 10123. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-662-54110-4_6
Download citation
DOI: https://doi.org/10.1007/978-3-662-54110-4_6
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-662-54109-8
Online ISBN: 978-3-662-54110-4
eBook Packages: Computer ScienceComputer Science (R0)