Usage Patterns and the Economics of the Public Cloud

Networks, the Internet, and Cloud Computing and Cloud Computing

Article Snapshot


Aadharsh Kannan, Cinar Kilcioglu, R. Preston McAfee and Justin Rao


WWW 2017: Proceedings of the 26th International Conference on the World Wide Web, May 2017


Airlines raise and lower prices to ensure that supply and demand are roughly equivalent. But cloud computing service providers usually do not use “peak load” pricing. So far, cloud providers rarely vary prices with demand because datacenter traffic loads are stable.

Policy Relevance

In the future, datacenter traffic is likely to be more volatile. Cloud providers would then benefit from using “peak load” pricing.

Main Points

  • Traditionally, firms would maintain their own information technology (IT) equipment such as personal computers and servers on their premises; now, specialist “public cloud” providers offer software and data storage over their networks as a resource shared by many firms.
  • Observers expect cloud computing to overtake the traditional model within ten years; firms benefit from using cloud computing in three ways:
    • The firm avoids large IT capital expenditures.
    • Firms use less or more cloud services as needed, paying only for what they use.
    • Firms avoid being “locked-in” to software.
  • In electricity, airlines, and hotels, capacity to supply the service is fixed and relatively costly, and providers use peak load pricing (also known as dynamic pricing) to ensure that demand does not exceed supply, and that supply does not exceed demand.
  • Cloud computing also has fixed, costly capacity, and some might expect cloud service providers to use dynamic pricing; however, puzzlingly, cloud service prices rarely vary with demand.
  • Data from a major provider’s public cloud data centers shows that, surprisingly, cloud service providers would not benefit much from dynamic pricing.
    • The majority varied no more than 5% in their usage of cloud services at the individual level, though a few were more volatile.
    • From the standpoint of the cloud service provider, fluctuations in customer’s usage “washed out,” and, overall, data center usage was stable and predictable.
  • The relative stability of data center usage explains why cloud service providers have not adopted “peak load” pricing.
  • Firms’ actual intensive use of computer processing resources varied as much as 30%, dropping during the weekend and at certain times of day; in the future, as more and more firms transition to cloud services, demand for the public cloud will tend to be more volatile; service providers would then benefit more from “peak load” pricing.

Get The Article

Find the full article online

Search for Full Article