3. Adopt server virtualization. Virtualization and Moore's Law: servers in data centers are like dandelions in a yard. At first you see a few, and then overnight the yard is full of them. Many data centers today have underutilized servers. By underutilized I mean used less than 7 percent during a 40-hour work week. About 10 to 30 percent of the servers in most data centers fit into this category. Conduct an inventory on which of these servers can be virtualized onto another new server and turn those older servers off. Also, how many servers do you have that are older than six years?
Moore's law states that every year processing power and efficiency get twice as good. It is time to either upgrade older servers to more efficient ones or virtualize them onto other servers. This can really add up to some major savings. For each server removed from the data center, you can easily save hundreds of dollars in powering and cooling costs. Virtualization is a sound and true technology that efficiently takes advantage of the horsepower in today's servers. It works great and should be seriously considered when greening up your data center.
4. Exercise power management. Most systems administrators are focused on performance and uptime for the data center. Many have had problems with power management utilities in the past. It is time for systems administrators to take another look at power management tools. All of the major CPU manufacturers have implemented stepping technology that scales back voltage and clock frequency. This can result in huge savings for power and cooling. I have seen a CPU running at 50 percent utilization easily save as much as 65 percent of power. That is an amazing gain from simply turning up a piece of free software.
And for systems administrators who are focused on uptime/performance, I would like to point out that power management reduces the power, heat, and cooling in a data center, which in turn increases reliability. I recommend testing this out with a few servers to see the results. I have seen some Linux-based servers get a little squirrelly with power management, so test to be sure. Also, turn the lights off in your data center. Servers are not afraid of the dark and can easily be managed remotely.
5. Optimize your power supplies. The single biggest waste of power in most data centers is the power supply. More than likely, this is the single biggest source of power waste in your data center. Let's say your server needs 300W to run, and you have a 600W power supply installed. That means that half of your energy is being wasted and burned up as heat, which now needs to be cooled.
A lot of commodity servers become commodity servers because the manufacturer tries to save money on the power supply. The vast majority of servers in data center today have power peaks at 75 percent with a 100 percent load and 65 percent with a 20 percent load. What? 10 percent is all the power I save? That is correct. A high-efficiency power supply should be able to deliver a consistent energy peak across 20 percent, 50 percent, and 100 percent loads. These power supplies start at an 80 percent efficiency rating and go up from there. They cost about 10 to 20 percent more than inefficient power supplies, but their payback is immediate. For example, if you spend $20 extra on a high-availability power supply in one year, you can add another $150 back into your budget. Now count your servers and look at the savings.
As a final step, review your electric bill after six months. I believe you will be very surprised and happy at the savings you have achieved.
Jimmy Ray Purser is a networking expert at Cisco.
Previous 
