Energy and power costs continue to command the lion’s share of the budget in data centers of all sizes. Data center facilities with 24x7 operations are among those with the highest density of energy-consuming equipment of any modern building.
“They can use 100 times the electricity of a typical office building on a square-foot basis,” said William Tschudi, energy analyst at Lawrence Berkeley National Laboratory. “Energy costs of $1 million per month are not uncommon in large data centers that require megawatts of electricity.”
Few organizations are more concerned with data center efficiency than Google, which requires one of the most powerful computing infrastructures in the world to conduct its operations. Google runs more than a million servers in data centers around the world, and processes more than a billion search requests and about 24 petabytes of user-generated data every day.
While Google’s research into data center efficiency includes such experiments as seawater-cooled data centers and solar thermal receivers, the company says the real key to efficiency is a rigorous application of best practices. In a recent company blog, Google identified several of its best practices and design considerations that can be applied to both small and large data centers.
Measure Power Usage
You can't manage what you don’t measure, so it is important to characterize a data center's efficiency performance by measuring energy use. Google uses a ratio called PUE — Power Usage Effectiveness — to help reduce energy used for non-computing activities such as cooling and power distribution. To effectively use PUE it's important to measure often — Google samples at least once per second. It’s even more important to capture energy data over the entire year because seasonal weather variations have a notable affect on PUE.
Manage Air Flow
Good air flow management is fundamental to efficient data center operation. Start with minimizing hot and cold air mixing by using well-designed containment. Eliminate hot spots and be sure to use blanking plates for any unpopulated slots in server racks. Google says thermal modeling using computational fluid dynamics (CFD) can help organizations quickly characterize and optimize air flow for a facility without disruptive reorganizations of the computing room. Also be sure to size your cooling load to your expected IT equipment, and if you are building extra capacity, be sure your cooling approach is energy proportional. .
Use Free Cooling
“Free cooling” is removing heat from a facility without using the chiller. This is done by using low-temperature ambient air, evaporating water or using a large thermal reservoir. Chillers are the dominant energy-using component of the cooling infrastructure, and minimizing their use is typically the largest opportunity for savings. There is no one right way to free cool, but water- or air-side economizers are proven and readily available.
Adjust the Thermostat
Raising the cold aisle temperature will reduce facility energy use. Don't try to run your cold aisle at 70F; set the temperature at 80F or higher —virtually all equipment manufacturers allow this. For facilities using economizers, running elevated cold aisle temperatures is critical as it enables more days of free cooling and more energy savings.
Optimize Power Distribution
Minimize power distribution losses by eliminating as many power conversion steps as possible. For the conversion steps you must have, be sure to specify efficient equipment transformers and power distribution units (PDUs). One of the largest losses in data center power distribution is from the uninterruptible power supply (UPS); be sure to specify a high-efficiency model. Also keep high voltage as close to the load as feasible to reduce line losses.
