03:45 PM
5 Tips to Achieving an Energy-Efficient Data Center
A 50 percent reduction in data center power consumption is a reasonable goal, according to independent technology consultant Neal Nelson. But, he says, it is best achieved through a series of small changes, which are outlined below:
1) An essential first step is to measure total power usage before an energy conservation program starts and continuously thereafter. This is usually done by installing submetering technology. The technology chosen must be able to track changes as little as 2 to 3 percent. Think about metering the power to the servers plus the heat, light and air conditioning.
2) This is not a fast process. Be prepared to make some changes and then collect data for at least several weeks to determine the impact of the changes.
3) Plan on at least three phases for the energy conservation program. Phase One should be simple changes that don't really cost anything to implement but will result in measurable power savings -- for instance, adjusting the thermostat in the server data center, turning off lights in machine rooms, manually powering off servers during periods of low demand and relocating servers to improve cooling efficiency.
4) After demonstrating that there are measurable savings from the first phase, Phase Two, which will require some investment to implement, can be more easily justified. Changes such as adding RAM to servers to improve disk caching and increasing throughput of individual servers can allow other servers to be powered off. Or replace several older machines with a single newer/faster server. An additional 20 percent savings could result from this phase. Document the savings with the submetering system and prove to management that the funds were a good investment.
5) Phase Three involves major changes (and significant costs). Examples include reconfiguring the data center with improved cooling (hot and cold aisles), installing a data-centerwide direct current power supply, and installing administrative software tools that power servers up and down in response to real-time workload changes.