IT equipment requires significant amounts of electricity. And, while servers today are much more energy efficient than older models, data centers need even more power to address the heat produced by IT equipment. To make matters worse, power capacity may be limited or inconsistent in some parts of the country. But there are strategies for making better use of the available power – and even options for data centers interested in generating their own power.
Download for later:
- Internet Explorer: Right Click > Save Target As
- Firefox: Right Click > Save Link As
In this podcast, Nick Martin, assistant site editor for SearchDataCenter.com, speaks with Bob McFarlane, a principal in charge of data center design for Shen Milsom & Wilke, to find out more about energy re-use and generation, and to discuss whether recent experiments to take data centers off the grid entirely are actually practical.
Nick Martin: First, what are the differences among cogeneration, independent power production and heat reclamation? Are these terms interchangeable?
Bob McFarlane: They’re used interchangeably by a lot of places in the industry, but they really should not be. Cogeneration is technically the production of two forms of usable energy from one fuel source. It's also known as “CHP,” or combined heat and power. The most common example would be a gas-powered turbine generator that generates electricity and heat. The electricity powers equipment and the heat is then used to run an absorption chiller to provide cooling.
Independent power production uses a local fuel source to run a generator with sufficient capacity to run the IT equipment. The resulting heat may or may not be utilized. So when people refer to this as cogeneration, it really isn't. The excess power can be sold back to the utility at the utility’s unused fuel costs. That’s the federal definition, and it is only a few cents per kilowatt. So if you're buying energy at 12 to 15 cents per kWh, for example, the power company may only pay you back three cents per kilowatt hour. So it's not a major moneymaker in most cases.
The efficiency gained by doing independent power reduction is based on two factors. One is running your generator at close to peak capacity. Data centers use about the same amount of power virtually 24 hours a day. You can size your generator to run close to capacity, which makes it very efficient. The utility can’t do that because they have the size for peak times, like the middle of the summer. So they're running at a relatively low efficiency most of the time. Also, you're avoiding transmission losses. A lot of power is lost to the wires and transformers as electricity travels miles and miles across towns. You're sending it just a matter of feet from generators to your equipment, so you're eliminating a lot of those losses. Independent power production can be quite efficient even if you're not selling power back to the grid at any significant gain.
Heat reclamation is simply reusing the heat produced by any energy-consuming equipment, capturing that heat and transferring it to useful purpose. Usually this is used for things like heating the building in winter and cooling in the summer by using an absorption chiller. However, it takes a lot of heat energy to run an absorption chiller, so that becomes a problem.
Martin: Which would be most practical for data centers to implement, and why?
McFarlane: All three are possible but with significant caveats. Independent power production is straightforward. It just requires a suitable generator plant. In fact, the infamous 2007 EPA study of data center power consumption strongly recommended that data centers produce their own power to reduce the need for the construction of new power plants.
Heat reclamation however, is probably the most common. It is used for things like keeping battery rooms at the right temperature or heating ancillary areas of a data centerthat can get pretty cold if they're being cooled by the air conditioners that are cooling the data center.
Cogeneration can be difficult. Large absorption chillers, contrary to popular belief, require more heat than comes out of the computing equipment. We think our data centers are becoming hot, but there is nowhere near enough heat to run an absorption chiller. Even most diesel generators don't put out enough heat to run a decent-sized absorption chiller. There are special chillers that run at these low temperatures, but these aren't really what we want for high-efficiency power generation. So cogeneration really requires a fairly large-scale generation to become attractive.
Martin: What are the major concerns when considering cogeneration or independent power production options?
McFarlane: Quite frankly, the biggest obstacles for both are environmental restrictions and return on investment. Most municipalities are not going to let you run conventional diesel generators continuously, except in emergencies. Large generating plants require operating personnel and potentially complex controls if it's large enough to power a data center. Given the small rebate on this sold-back power, cost reduction may never occur. Now if you're out in the boondocks somewhere, where we don't have to worry too much about noise pollution, then you may be able to run a large generating plant. Gas turbine generators produce a different kind of noise that can be easier to control, but you still have to deal with all the pollution regulations. It's almost all environmental.
Martin: Do you see independent power production technologies emerging or becoming more mainstream in the years ahead?
McFarlane: It depends on the type of power production. In 2009, the Uptime Institute made generators a requirement for all of their infrastructure tier levels. If you're designing with the Uptime Institute’s levels as any kind of a guideline, then all tiers require generators.
The Uptime Institute added another very interesting statement to their tier levels recently. They said to qualify as a Tier 4 facility, the generators must be considered the primary source of power, with the public utility considered to be a cost-effective alternative that is expected to fail. This clearly says that local generation should become more mainstream, if you can get around the environmental restrictions.
Now we have a whole different kind of situation if we go to things like solar panels, wind turbines and geothermal. These all limit the pollution problem, unless you consider visual pollution a concern. The problem with these alternatives is efficiency and capital costs. It takes a lot of solar panels or wind turbines to power a data center, and there's no giant breakthrough on the horizon that I'm aware of. There are also technical issues in coupling these things to the grid if you want to sell power back to the public utility. Of course, if the sun is shining or the wind stops blowing you still need the grid to keep running. The real answer in the long term may turn out to be fuel cells. They produce both power and substantial heat without relying on nature, but they’re still very expensive and they have limited availability in the sizes we need for data centers. People are watching and designers are trying to build better fuel cells. In my opinion, that may turn out to be the ultimate answer for our industry.
Martin: Will data centers ever be self-sustaining and "off the grid" entirely?
McFarlane: The EPA would love that. It takes real guts to disconnect completely from the grid, and if you stay connected but never use it, you’re going to be charged mightily for the privilege -- the utility is not going to be interested in putting big power systems into a facility and letting them sit there unused while you don't pay for any power.
There is an experimental data center project going on in New York and Texas that is going to go completely off the grid. It intends to rely solely on the wind and solar generation with a lot of wind in upstate New York and, of course, a lot of sun in Texas. But they’re building two data centers in very different locations with the expectation that, if one runs short of power, they can shift complete computing operations to the other site. This is a different approach, and I think you have to say it's a rather expensive form of redundancy. Still, there's no guarantee that conditions might not be unfavorable at both sites simultaneously. They still believe they can do it with no connection to the grid entirely, and only time will tell.