The Real Cost of a CPU Hour

Edward Walker
2009 Computer  
Published by the IEEE Computer Society 0018-9162/09/$25.00 © 2009 IEEE IT organizations can now outsource computer hardware by leasing CPU time through cloud computing services. A proposed modeling tool can quantitatively compare the cost of leasing CPU time from these online services to that of purchasing and using a server cluster of equivalent capability. T he acquisition and installation of IT server clusters is growing significantly worldwide. 1 Partly fueling this growth is society's
more » ... h is society's reliance on the Internet for collaboration, entertainment, and communication; the adoption by IT organizations of the concept of software as a service; and the use of simulation in scientific discovery and product development. At the same time, commercial companies like Amazon.com, IBM, and Google are now letting any organization, indeed anyone with a credit card, purchase time on servers in their data centers through online Web services. With these "cloud computing" services, IT organizations can lease only the compute time they require for their computational needs, rather than purchase a server cluster. Since August 2006, for example, Amazon.com has offered the Elastic Compute Cloud (http://aws.amazon. com/ec2) Web service. For $0.10 per CPU hour, organizations can purchase virtual machine instances with root access, running on a compute node in an Amazon.com data center. What does $0.10 per CPU hour mean? Is this cost fair? How does this cost compare to that of purchasing a server cluster? Providing quantifiable answers to these questions is of intrinsic value to computational scientists, IT organizations, and equipment-granting agencies. To allow consumers to make rational choices, a tool is needed to quantitatively compare the cost of purchasing a cluster versus the cost of leasing CPU time from an open competitive market. My proposed model computes the real cost of a CPU hour when performance depreciation (the time value of a CPU) is taken into account. Deriving this cost enables a quantitative comparison of the investment choices of leasing online compute time versus purchasing a server cluster. Understanding this tradeoff through modeling lets IT organizations justify purchase decisions quantitatively. In addition, funding agencies can objectively evaluate alternative equipment proposals, and policymakers can fashion guidelines that ensure the most efficient outcome for all. Most importantly, if consumers understand the real cost of a CPU hour, an efficient market will eventually reflect this real cost, enabling a fair, competitive market for the benefit of all consumers. CPU LEASING ImPACt Acquiring assets is a central challenge in any organization, and deciding whether to lease or buy equipment isn't a new dilemma in economic finance. In fact, more than four decades of research describes models for rationally making equipment lease-or-buy decisions. [2] [3] [4] An often used decision model calculates the net present value (NPV) of
doi:10.1109/mc.2009.135 fatcat:7nznzmfyozcupcp44gwskktthq