- Sunday, 29 November 2009
- INQ Mobile
- RSS
You call it moving the goal posts -- I call it dynamic e-business
Intel tries using normal outside air to cool datacentres
INTEL JUST RELEASED a white paper which reckons servers can stick it out in pretty tough conditions, cooled only with outside air and save money to boot.
The Chipzilla report on "free cooling" talks about using something called air-side economisation which sucks in outside air to cool datacentres, pushing the hot air from the machines outside. The simplicity of it sounds stunning but even more stunning is that it could mean the end of the expensive, energy sucking CRAC units currently being used.
Testing to see if machines could actually be cooled by outside air temperatures as high as 32 degrees Celcius, Chipzilla also decided to challenge the notion that outside, unpurified air was bad for servers. Turns out, it didn't do them too much harm.
Setting up a datacentre in New Mexico, with approximately 900 used production servers (standard, single-core, dual-socket, Intel-based machines) in a 93 square metre trailer – divided into two parts – Intel cooled one part with a warehouse-grade DX air-conditioning unit and the other almost only with fresh air.
When the outside air temperature reached over 32 degrees C, Intel switched on the aircon, and when temperatures dropped to below 18 degrees C, Chipzilla warmed the air up a bit by mixing outside air with hot air coming out of the server.
Over the 10-month test, Intel figured its approach, whereby air economisers were used about 91 per cent of the time, saved about 67 per cent of the power needed for a 10MW datacentre, translating into monetary savings of $2.87 million. Cool indeed.
And as for damage to the machines by humidity and bits of contamination? Well, considering no humidity controls were used whatsoever, and only a bog standard household air filter was put in, the results were not unimpressive.
"There was only a minimal difference between the 4.46 per cent failure rate in the economiser compartment and the 3.83 per cent failure rate in our main datacentre over the same period," noted Intel. Also, "the failure rate in the trailer compartment with DX cooling was 2.45 per cent, actually lower than in the main datacentre," according to the white paper.
Of course, with humidity fluctuations from four to over 90 per cent for machines cooled by outside air, the results were far from perfect, but at 74 per cent fewer KWhs to cool the trailer, Intel shouldn't let a bit of humidity dampen things. µ
L’Inq
Intel's
white paper
Video on how the air economiser works
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
For an in-depth presentation of optimizing HVAC (heating, ventiliation, and air-conditioning), see "Optimization of Unit Operations", by Bela Liptak. One of my process control bibles.

Sounds like Intel actually hired a real process control engineer. Good for them.

But nothing new at all, the same principles can and are applied to building HVAC system. Somebody made the miniscule leap to figure out this would work for data centers, also. I'm glad the computer industry is SO creative.
I think that in the future, big datacenters will be placed up in high elevation mountains where the air is cold even during the middle of summer.
Intel picked the site because the climate is extremely dry. Albuquerque averages about 8" (400mm) of rain/snow and < 20% relative humidity and is at 1700-2000m elevation with excellent conditions for radiational cooling. That is not the real world. So unless everyone is going to relocate server farms to places like the Chihuahua Desert or the Tibetan Plateau this smacks more of publicity stunt than reality. Mold and mildew would kill this in most climates.
in the standard data center failure rate 3.x%
in the DX cooled center failure rate 2.x%
in the ventilated center failure rate 4.x%

DX center and ventilated center were using end-of-life equipment.

Production datacenter was using new equipment

i.e. new equipment will have a roughly 3.x% failure rate.
after three years, the probability of failure decreases to 2.x%.
If you do not properly cool your equipment, there is a 2x greater chance of equipment failure as demonstrated with the end-of-life equipment test.

Therefore, if the new equipment were subjected to the same conditions, you can probably expect a 6.x or greater failure rate.

Is a >6% failure rate in your datacenter acceptable?

Have a nice day!
Living in Colorado I have always wondered about the need for AC units in data centers and labs to cool down all the computers. I think it is a great idea for something like this. Now if only companies would start implementing it.

Also wonder if they could just dehumidify the air in certain places - but this might be done with AC units I don't know. In CO we probably wouldn't need to since our air is pretty dry anyway.