Cover V10, I09

Table 1
Table 2
Table 3


Why Energy Consumption Matters

Gilbert Held

Serious energy problems are facing residents of California and other western states. Commencing with utility companies reducing voltage by a few percent during the summer of 2000 that resulted in brownouts, the electrical situation rapidly deteriorated to rolling blackouts during the beginning of the year 2001. With the summer of 2001 upon us, it is apparent that both business and residential electrical consumers will continue to face a challenging situation. What may not be as apparent is that, as the cost of energy increases, so does the need to examine the operational energy cost of computer equipment during the equipment acquisition cycle. In this article, I will review the manner by which energy is billed and explain why a small change in the cost per kilowatt hour (kwh) can significantly affect the operating cost of data center equipment. I will examine the energy consumption of several general types of equipment and note how a small difference in the consumption level can affect the lifetime cost of equipment.


Consider the three key components that define an electrical circuit. Those components are the current flowing through a circuit, the voltage potential used to drive the current through the circuit, and the resistance to the flow of electrons in the circuit. The relationship between current (I), voltage (V), and resistance (R) is given by Ohm's Law where:

I = V/R or V = R x I

The resistance unit is the ohm, while current and voltage are expressed in terms of amperes and volts. As current flows through a circuit, heat is generated based upon the current flow and resistance. This heat represents the power expended due to resistance and is expressed as:

P = IV = I2R = V2R

where P is expressed in watts. Because power is formally defined as the rate at which work is performed, we would measure power expended during a time interval (1 hour). Thus, a watt-hour is used to define work performed by a device at a constant rate of 1 watt for a period of 1 hour. Because the watt is inconveniently small, the kilowatt (1kw = 1000w) and the megawatt (1mw = 1000kw = 1,000,000w) are more commonly used. A kilowatt-hour (kwh) is the unit by which an electric power company typically bills its consumers. It should be noted that a kwh technically represents a unit of work or energy and not a unit of power. Thus, the bill we receive each month for consuming a certain number of kwh is for energy consumed.

KWH Cost Variations

The cost per kwh varies considerably from state to state and even within a state based upon a number of factors. Those factors include the cost of fuel used to generate electricity, the distance of a power plant from the location of the fuel, property taxes the utility pays to the local community, and a potential range of taxes the ultimate consumer finds added to their monthly bill. The end result is a cost per kwh that varies from approximately $.05 per kwh in the Pacific Northwest where hydroelectric power generation represents both a renewable and economical source of electric generation, to over $.10 per kwh in the Northeastern portion of the United States where generators primarily depend upon imported oil.

24/7 Operations

When I was growing up, my parents followed me around the house turning off unneeded lights. Although we cannot walk around a modern data center turning off servers, we can appreciate the need to compare the operational costs of different devices.

Most devices in a modern data center operate almost continuously, being turned off only for an upgrade, repair, or planned maintenance. Equipment such as routers, hubs, access controllers, gateways, and Web servers are normally referred to as 24/7 devices, indicating that they operate all day, each day of the week. To illustrate the energy operational cost of a 24/7 device, let us assume you install a small server which typically consumes 250 watts of energy. On an annual basis, excluding leap years, there are 8760 hours. Thus, the server would consume 2,190,000 watt hours in a year. Because utilities normally bill based upon kwh consumed, the electrical bill would reflect the use of 2190 kwh.

If we assume the server is located in the Pacific Northwest, we might be billed $.05 per kwh or $109.50 annually to keep the server operating. However, several utilities in the Pacific Northwest had to import energy and raise their rates between 20-40 percent.

At a more common rate of $.10 per kwh, operational cost of the server doubles to $219 annually. The operational energy cost at $.05 per kwh or $.10 per kwh may not appear significant on an individual server basis for one year, but consider the rest of the story. To obtain a bigger picture of the operational costs in the data center we need to examine the cost of operating each device over its expected life. For example, assume the server has an expected life of 4 years. Then, at a cost of $.05 per kwh, the energy operational cost of the server would be $109.50/year x 4 years or $438; at $.10 per kwh, the cost would be $876 over 4 years. In fact, every penny per kwh increase or decrease results in a change of $21.90 per year, or $87.60 over the expected life of the server.

Note that the server for which we just computed several basic energy operational costs would probably be classified as an entry-level product. Thus, the one-time cost of this server (to include perhaps 50 to 100 GB of online storage) would be approximately $3000. Table 1 summarizes the energy operational costs for an entry-level server and includes the one-time cost of the server for comparison.

In Table 1, note that the expected 4-year operational energy cost approaches one-third of the purchase price of a low-end server. Consider the increases in wholesale power prices during the first few months of this year. During February 2001, California energy costs averaged over $300 per megawatt hour at the wholesale level. This is equivalent to $.30 per kwh and perhaps provides the key reason why it is probably a matter of foregone conclusion that electricity rates can be expected to increase. As they increase, the cost of operating 24/7 devices within a data center will represent an ever growing percentage of the purchase price of data center equipment.

Facilitating the Cost Computation Process

To facilitate the computation of the annual cost associated with 24/7 always-on equipment, a spreadsheet was used to determine the relationship between watts consumed and cost per kilowatt hour on the yearly operating cost of equipment (Table 2). The intersection of a column and row in Table 2 indicates the annual cost for an always-on device based upon its level of energy consumption and cost per kwh. Through the use of this table, you can easily determine the annual operational energy cost for devices that have a consumption range outside of the wattage column or beyond the cost per kwh. For example, assume your new server consumes 3700 watts and will be installed in a data center where the cost per kwh is $.12. Note that there is no $.12/kwh column nor is there a 3700 watt row. Thus, you would first move along the 1000 watt row until it intersects with the $.10/kwh column and note the annual energy cost is $876. Multiplying by 3 results in an annual cost of $2528 for 3000 watts at a cost of $.10/kwh. Noting that the operating cost for 700 watts consumed is $613.20 on an annual basis at a cost of $.10/kwh, for a total cost of $3,241.20, again at a cost of $.10/kwh. Now all we need to do is multiply $3241.20 by 1.2 to compute the operational energy cost at $.12/kwh instead of $.10 per kwh. Doing so results in the operation energy cost becoming $3,889.44 when our organization's bill for energy consumption is based on $.12/kwh for a high-end server with RAID disks that consume 3700 watts. See Table 2.

Examining Equipment

If you carefully examine your data center equipment, you may notice a label on many devices stating their power supply value in watts. That value typically represents the maximum energy load the power supply will consume, but not its steady state rate of energy consumption. Engineers designing equipment typically use a power supply that is between 20 and 30 percent beyond the device's steady-state condition. However, before celebrating that the actual cost to operate your server farm may be less than expected, you also need to consider the heat dissipation of your equipment.

Every device that consumes energy generates heat, which explains the need for fans in many types of computer and communications products. The literature for routers, gateways, access controllers, and servers may contain a heat dissipation figure for each device. That heat dissipation is given in terms of British Thermal Units (BTUs) per hour, where the BTU was originally defined as the energy required to raise the temperature of one pound (.45 kg) of water one degree Fahrenheit (F) from a temperature of 59.5° F to 60.5° F. (There is now a more formal definition because the amount of energy required to heat water differs based upon the current temperature of water.)

You or your air-conditioning engineer probably had to compute the BTUs generated by equipment when the tonnage of the building's air conditioner was sized. Thus, the heat generated by servers, routers, gateways, and other devices must be removed from the data center. Due to this, the air-conditioning system has to work harder, which increases the cost of maintaining the building.

Instead of becoming immersed in computations, simply consider the cost of operating each device and the cost of removing heat generated by the device by using the heat dissipation value. This simplification works well for most locations, since during the winter the extra heat can be recirculated in many buildings. However, if your data center is located in a hot and humid location, you may need to reexamine the heat dissipation as a separate entity.

Considering a Total Acquisition Cost

Until recently, when we evaluated the acquisition of such equipment as routers, gateways, and servers, we focused on functionality and purchase price. If we examined products from two or more vendors and only one vendor provided the functions and features required by our organization, life was simple and we would justify what was a sole-source procurement request. If two or more products satisfied our requirements, we would then turn our attention to the one-time cost of the equipment, assuming we were only considering purchasing and such additional costs as shipping and annual maintenance. Because most data center managers might have a single line item for utilities, rarely did organizations use the operational energy cost as a factor in considering products from different vendors. However, over the past few years, the growth in the use of the Internet has resulted in server farms that may consume more than 4 percent of all electricity generated in the United States. When you add in the basic Internet infrastructure of routers, hubs, access controllers, etc., conservative estimates place the Internet as responsible for more than 6 percent of all energy consumed. As energy consumption increased, the not-in-my-backyard (NIMBY) syndrome against building power-generation plants resulted in an energy surplus migrating into a shortage in many locations, with the cost per kwh beginning to significantly increase. As the cost per kwh jumps by 20 to 30 percent or more annually, it now becomes important to consider the operational energy cost of products as they are being acquired. Thus, I will now focus on how we can use the operational energy cost as decision criteria in an equipment-acquisition process.

For that sake of simplicity, we will use the old illustrative standby of Vendor X and Vendor Y for comparison purposes, comparing the total cost of ownership of a medium-range server from each company. We will also assume that both vendors provide equipment that satisfies our basic organizational requirements. Assume we examined the financial aspects of each server and recorded relevant information, as indicated in Table 3.

Based upon the initial financial comparison of servers from Vendors X and Y, it appears that Vendor Y provides a lower total cost of ownership, but does it? Table 3 does not reflect the cost of operating the equipment. Assume we contacted each vendor and determined that the server from Vendor X consumes 2800 watts while the server from Vendor Y consumes 3300 watts and the cost of electricity is $.10 per kwh. What affect does this difference in energy consumption have upon the total cost of ownership of each server?

At an energy consumption level of 2.8 kw, the server from Vendor X consumes 2.8 kw x 8760 hours/year, or 24528 kwh per year. At a cost of $.10 per kwh, the operational energy cost to keep the server from Vendor X ticking on a 24/7 schedule would be $2,452.80. For the server from Vendor Y, the consumption of 3600 watts results in the consumption of 3.6 kw x 8760 hours/year, or 31536 kwh. Using a cost of $.10 per kwh, the annual charge to keep the server from Vendor Y running would be 31536 kwh x $.10/kwh, or $3,153.60. On a 4-year life cycle, the cost for the server from Vendor X becomes $2,452.80/year x 4 years, or $9,811.20, while the 4-year cost for powering the server from Vendor Y becomes $3,153.60/year x 4 years or $12,614.40.

If we now add the energy operating cost for each server to the total cost shown in Table 3, we obtain a more valid life-cycle cost. Thus, the Vendor X life-cycle cost becomes $33,411.20, while the Vendor Y life-cycle cost becomes $35,084.40. Note that although it initially appeared that the server from Vendor Y was a more economical purchase, when we considered the energy consumption level of each device, Vendor X had a lower life-cycle cost. While the 4-year difference is only $1,673.20 in this example, if you were acquiring several servers, the cost difference would become more meaningful. Additionally, do you believe the cost of electricity will not continue to rise in the future? If you, like me, expect the cost to increase, then it is reasonable to consider that cost in your acquisition process.

Gilbert Held is an award-winning author and lecturer. Gil is the author of over 40 books and 400 technical articles to include Cisco Access List Field Guide and Cisco Router Performance Field Guide, both published by McGraw-Hill. Gil can be reached via email at: