Correlating the Cost of IT Emissions with Server Location

There are three main costs associated with running a server that are related to energy –  hardware power requirements, cooling, and housing infrastructure (space). The umbrella term of "Green IT" has until now been associated mainly with reducing these operational costs associated with the environment. Purchasing more efficient IT requires investments, but these investments usually pay off with noticeable savings in terms of utility costs and other environmental factors. However, because the percentage of total US energy used by IT has increased rapidly, there is a possibility that new legislation will be introduced targeting the emissions of IT equipment. If the government starts regulating CO₂ use by IT, an enterprise will need to calculate the premium associated with drawing power generated by cleaner sources (currently electricity from green sources has an average premium of +1.64 ¢/kWh nationally), and that cost could vary considerably by location. The absolute CO₂ emissions incurred for delivering the power to a server will depend on the power infrastructure in its location. As a result, IT managers could soon have some new variables to consider as they determine where to locate their datacenters most cost-effectively.

A 2007 academic study showed the breakdown of power utilization by component in a server.

The energy efficiency of many of these components can be improved through hardware adjustments and server management software. These upgrades, coupled with the fact that computer performance grows yearly, means that it is possible to pursue sustainability through server software and hardware upgrades. IDEAS ServerCAR can be used to navigate through the mountains of data that one would need to compile in order to achieve these goals (updating servers, consolidation simultaneously with power savings, and reduction in CO₂ output). To gauge performance, ServerCAR uses the IDEAS Relative Performance Estimate 2 (RPE2) data, which ranks servers based on performance information that is distilled from published benchmark figures. Importing a user’s collected inventory list from Excel, ServerCAR intelligently maps servers down to a configuration level and can then set a target for performance growth.

For example, suppose a company installed two IBM System x3200 Quad Core servers with 2.13GHz processors in their datacenter in 2007. If the enterprise was satisfied with the IBM brand, and just wanted to upgrade the server to the System x3200 M3 Quad Core with 2.67GHz processors, a 20% performance growth would be achieved with just one server (reduction of 50%). Looking at the green metrics, this consolidation drops the max power consumed from 1,050 to 523 Watts. This represents an annual savings in New York of $1,341 (USD). With the dip in electricity usage, there is a corresponding dive in CO₂ output per year. According to ServerCAR, with the new server, the enterprise is dropping the total tonnage of CO₂ output by 57%.

 

 

Carbon footprint drop for upgrading from IBM System x3200 Quad Core servers to System x3200 M3 Quad Core

There is also a savings in cooling cost (3,588 BTU vs. 1,784 BTU with the new hardware).

However, another factor that a business might consider in addressing their carbon footprint is the geographical location of the server. Physically separating computing resources from their access points allows an enterprise to consider regional advantages in terms of power costs and emissions. Additional variables could then possibly be added to the equation, including legal, latency and security issues, as well as differences in taxes, labor cost, electricity cost, and CO₂ output. All of these factors can make the task balancing cost and environment challenging.

With public utilities, it is possible to mitigate CO₂ output by buying clean energy at a premium. If the same large company wanted to dedicate a percentage of this energy to clean sources, the CO₂ output would decrease, but their electricity bill would increase. At the same time, the servers could be moved to another location where the clean energy premium is lower, but the overall electricity costs are higher.

By combining ServerCAR output and publically available data about renewable energy premiums, it becomes possible to compare three specific issues: electricity costs, CO₂ output, and the premium for using power with substantially less C0₂ output . By outputting the data from all US states from ServerCAR (the figure below is a small representative sample), it is possible to compare CO₂ emissions in tons (the area of the circles) and price vs. clean energy premium (center of the circle) for a server model in each state. As shown in the figure, with all those other variables mentioned being equal, a large company that is not limited by geography, or latency concerns, would target a smaller circle close to the origin (i.e. the smallest cost and premium).

 

 
System x3200 M3 Quad Core server environmental costs - clean energy premium (¢/kWh) vs. power costs (USD) vs. overall CO₂ footprint by state (tons)

A few observations follow from this exercise. First, given the low absolute costs of power, high rate of clean energy usage overall, and a corresponding low premium for use of clean energy, it is easy to see why Washington and Oregon are considered the ideal locations for large datacenters, such as cloud computing centers constructed by Microsoft and Google. Running the System x3200 M3 in Massachusetts (MA) has an annual electricity price tag of $1,498 (USD) and 5 tons of CO₂ emissions, while the same server in Washington would only have an annual electric bill of $552 (37% of MA) and an annual CO₂ output of 1 ton (20% of MA). 

However, a few other anomalies become visible as well. In terms of yearly electricity cost, Arizona and Ohio are roughly equal, but the story changes completely if a company there wanted to cap their emissions. Ohio, currently the bigger emitter (larger circle), has a small premium for purchasing green energy, and so the amount of CO₂ output could be trimmed with a small effect on annual electricity cost. This action corresponds to a shrinking of the circle and an upward shift in on the y-axis. In Arizona, however, the amount of CO₂ output is less to begin with (smaller circle), but the marginal cost of lessening the carbon footprint is much greater because of the premium. Making the approximation that green energy produces no CO₂, the cost to shrink the circle to an area of zero would correspond to a new electricity bill of around $880 in Ohio and $1180 in Arizona. For states that started out equal that is a large difference!

Even with a non-exhaustive analysis of the data one can draw many conclusions. The landscape of future CO₂ legislation is uncertain, but with a little number crunching many definitive outcomes can be certified and companies can indeed make contingency plans.