The value of one watt of savings from more efficient IT equipment in a data center
In a white paper I wrote for Samsung last fall (Koomey 2012), I analyze the economics of purchasing green DRAM for servers. Assessing the economics of efficiency improvements in IT equipment requires knowledge not just of energy prices and energy savings, but also of the avoided infrastructure costs associated with lower power computing.
For new facilities, this is a real avoided cost, but for existing facilities that are capacity constrained it’s an opportunity cost–electricity that powers unnecessarily electricity-intensive DRAM chips could have been used instead to power another server that generates useful work. It’s only in facilities that are not capacity constrained (which are the vast minority, based on anecdotal evidence) where this opportunity cost is not relevant. There are large variations in those infrastructure costs depending on reliability requirements and type of data center, but it’s still important to understand the rough orders of magnitude for analyzing the economics of improving the energy efficiency of computing equipment in those facilities.
I summarize the results from these calculations in the following graph. The key result is that avoided infrastructure savings represent more than half of the economic savings associated with reducing computing electricity use in the data center (under plausible assumptions for the relevant parameters in different types of data centers). These economic benefits really matter, and analyses that don’t count them will mislead by substantially underestimating the benefits from efficiency improvements in computing equipment.
What’s one watt of electricity savings worth in the data center?
Notes: Infrastructure capital savings apply to new construction or existing facilities that are power/cooling constrained. Those savings total $8.6M/MW for cloud facilities and $15M/MW for others, from Uptime institute. PUE = 1.1, 1.5, and 1.8 for Cloud, New, and Existing data centers, respectively. Electricity price =$0.039/kWh for cloud facilities and $0.066/kWh for new/existing data centers. All costs in 2012 dollars.
Avoided infrastructure costs for cloud computing case reflect the lower bound of the 2nd quartile from Stanley and Schafer 2012. For typical data centers, avoided infrastructure costs reflect the median value from Stanley and Schafer. Avoided infrastructure costs only apply in existing facilities when they are power and/or cooling constrained.
Here are some more details on the Uptime estimates of infrastructure costs, summarized in the green DRAM report.
The cost of data center infrastructure is commonly summarized in millions of dollars per MW of IT load (which is equivalent to dollars per watt). According to data compiled by 451 Research (Stanley and Schafer 2012), conventional data centers range in cost from about $5M to $29M/MW (2012 dollars), with 50% of the data centers falling between $8.6M and $18.8M/MW. The median of their sample is at $15M/MW.
After conversations with the first author of the 451 Research report (John Stanley), I decided to use the bottom of the 2nd quartile ($8.6M/MW) as the avoided cost for Cloud Computing installations. Stanley was concerned that data centers with costs at the lowest end of the range ($5M/MW) might not be comparable in reliability or other important characteristics with the high performance data centers that make up the vast majority of the 2nd and 3rd quartiles of the distribution. I also decided to use the 451 Group’s median numbers ($15M/MW) as the value for Typical Existing and Recent Practice Facilities because “in-house” data center facilities are the ones most heavily represented in that report’s data sample.
So if you read or create analyses of the economic benefits of improving the efficiency of computing equipment, make sure they correctly account for the avoided infrastructure costs, which are substantial even in the most efficient data centers in the world.
References
Koomey, Jonathan G. 2012. The Economics of Green DRAM in Servers. Burlingame, CA: Analytics Press. November 2. [http://www.mediafire.com/view/uj8j4ibos8cd9j3/Full_report_for_econ_of_green_RAM-v7.pdf]
Stanley, John, and Jason Schafer. 2012. The Economics of Prefabricated Modular Datacenters. San Francisco, CA: The 451 group/451 research. May 11. [https://www.451research.com/report-long?icid=2266].
