X
Tech

Green tech expert clarifies data center heat discussion

Just wanted to pop in a quick update/observation from a reader regarding my "Some like it hot" post the other day. He wanted to clarify my reference to the Arrhenius equation, which I used after the Sun exec I interviewed for the story used it to illustrate what we were talking about.
Written by Heather Clancy, Contributor

Just wanted to pop in a quick update/observation from a reader regarding my "Some like it hot" post the other day. He wanted to clarify my reference to the Arrhenius equation, which I used after the Sun exec I interviewed for the story used it to illustrate what we were talking about.

Easier just to directly quote the comments in their entirety, which come from Peter Hopton, managing director of Very PC, a Green IT systems builder from the United Kingdom. Thanks, Peter, for caring enough to comment so thoroughly.

"The Arrhenius equation relates to the slow degradation of silicon over time, but the reason for immediate failure at high temperatures is due to a different factor called thermal runaway.

Arrhenius relates to the MTBF (mean time before failure) and suggests that a unit running at 20 degrees centigrade that lasts for 250,000 hrs will only last 125,000 hours at 30 degrees centigrade. Arrhenius would suggest that this server running at 50 degrees centigrade would be good for 31,250 hours.

You see, as temperature goes up in a semiconductor, the current flowing through the semiconductor increases, and the amount of heat produced increases. If this heat can’t be removed from the chip, it gets hotter still and produces more heat, until either a thermal cut off engages or the chip literally melts (and potentially causes fire). This is known as thermal runaway.

We already have servers that can run at 50 degrees centigrade, but the problem is: 1. Other equipment at 50 degrees centigrade: a. Switches, routers and other gear spurt out erroneous data at this temperature b. UPS equipment overheats and becomes much less efficient c. Resistance increases in communications cables degrading signal quality

2. Cooling inefficiencies start to outweigh heat-pump inefficiencies a. Fans have to pump air much harder to keep the ‘delta’ (change in temperature over the heatsink) down to the chips, these fans also produce heat when they operate

3. The thermal capacity of air is low a. So if there is a power cut, and the air conditioning goes off for 25 seconds while the generators kick in, the air will increase to 60 or even 65 degrees centigrade, and thermal runaway will occur in almost all equipment."

So, a much more complete explanation of why data centers currently need to be kept cool enough to account for worst case scenarios and why both systems vendors and data center operators are starting to question how to recalculate this equation.

Start your own dialogue with me on any green IT topic, by e-mailing hccollins@me.com.

Editorial standards