Next month is the one year anniversary of a guideline by the American Society of Heating Refrigerating and Air-Conditioning Engineers (ASHRAE) that recommend increasing (PDF document) the temperature of air entering servers and other data centre equipment. This increase of 77 degrees Fahrenheit to 80.6 degrees may not seem like a big deal, but it took a year-and-half of work to arrive at this recommendation and agreement by most of the major equipment vendors.
Cooling a data centre to 20 degrees Celsius may be going out of style.
The person who led the society's IT team on Technical Committee 9.9 was Roger Schmidt, an IBM fellow and its chief engineer for data centre energy efficiency.
It's unknown how many data centres have adopted the recommendation, or even have enough control over their environments to safely regulate air flows. Ken Brill, executive director of the Uptime Institute in Santa Fe, New Mexico, said he sees more understanding that the data centre temperatures can go up and says there has been a "very significant attitudinal change in a year," he said, but adds, that "many still don't know."
When evaluating power requirements during a data centre refresh, many organizations are too caught up with the big picture and miss the smaller factors that contribute to data centre heat output.
In an interview, Schmidt looked at the new temperature parameters, as well as some other issues involved in cooling data centres and reducing power usage.
How much heat can servers handle before they run into trouble?
The previous guidelines for inlet conditions into server and storage racks was recommended at 20 degrees Centigrade to 25 Centigrade. This is where the IT industry feels that if you run at those conditions you will have reliable equipment for long periods of time. There is an allowable limit that is much bigger, from 15 degrees Centigrade Fahrenheit to 31.6 degrees. That means that IT equipment will operate in that range, but if you run at the extremes of that range for long periods of time you may have some fails. We changed the recommended level -- the allowable levels remained the same -- to 17.7C to 27.2C. That means at the inlet of your server rack you can go to 27.2 degrees -- that's pretty warm. [The standard also sets recommendation on humidity levels as well.]
What made it possible to change the recommendation?
It took a year-and-half of all the IT manufacturers talking through this and making sure we had what we felt was some hard data behind this that would meet the new requirements.
Since this standard went out one year ago, what's been the adoption of it? Some are starting to use it. We [IBM] are starting to internally use it. It's something that's not going to happen overnight. They [data centre managers] will probably step it up two degrees at a time. The benefit will be contingent on an analysis for that data centre on what happens if you raise the air temperature and thereby raise the chilled water temperature by "x" amount. Raising the temperature allows you to possibly to turn off the chiller for a longer period time and use outside ambient air to cool your data centre. In general, it's like raising the thermostat in your house.
Do you feel is the 27.2 is a conservative upper recommended limit?
Above about 25 degrees we all start to speed up our [equipment] fans as the temperature gets higher in order to keep the silicon at a pretty level temperature. We don't want the chip temperature to be jerked around. As the temperature in the inlet into the rack goes up we speed up the blowers to increase the heat transfer, if you will, and to keep that silicon kind of constant. If you start to raise temperature more and more, the blowers and fans speed up more and more, using more power. This is not good. We feel the power increase is minimal for that level, but we did feel that raising it higher than that [the recommended limit] may end up diminishing returns for saving power at the whole data centre level.
What is the future of water, refrigerant-based cooling?
We got out of water cooling in 1995. In 2005, we announced the Cool Blue Rear Door Heat eXchanger that was applied to our high-end x86-type market because some of our products had very high heat loads that it created hot spots in data centres. This rear door heat exchange, which had cold water running through it, took out a major part of the heat load of the rack before it caused any problems for any other servers in the data centre. Since that time, we now offer that across all our x86 products, our high-end P Series products, our Unix-based products, and also have it on every sale we make on our Power 6, 575 high-performance computing product that's shipped out the door.
How ubiquitous will water cooling be?
We've gotten into pretty big time now especially with the Power 6, 575 bringing water down to the processor level. I think there will be more and more of push into this area. Education will be needed. Clients that live during the bipolar days [cooling technology prior to 1995] have no problem with this. It is two sets of clients in the market.
Is data centre location being driven by free air cooling? Can we expect to see a lot of data centres in colder climates?
Free air cooling is a big topic.Customers are out there locating data centres in climates that can use outside, I'll say, air conditioning. A concern are contaminates that you may bring in from the outside. Temperature fluctuations, humidity fluctuations: All those have to be monitored so you are in the requirements of what the IT equipment manufacturer says is acceptable.