• Forum has been upgraded, all links, images, etc are as they were. Please see Official Announcements for more information

Stonehedge is back!

Stonehedge, would peltier cooling devices slapped straight onto the walls attached directly to the cpu heatsinks work as cooling devices thus transferring the heat directly into the ground or would this an inefficient use of electricity. Another way might be a ground source heat pump in reverse?
 
It isn't really a question of could it work, more of a question of how could you do it? A fully loaded blade chassis is a very dense device with no room to gain entry to the CPU heatsinks without interrupting airflow. In fact, airflow is king when it comes to DC cooling. This is an extract from Cisco's data centre bible to give you an idea of the mainstream (and high budget) way of doing it.

Screen Shot 2014-10-20 at 23.05.40.png


As a rule, ambient air is sucked into the front of a rack and blown out the back hot. To cater for this you ned to ensure that you have hot and cold aisles within your DC. Your HVACs take in warm air at the top and blow it under the false floor where it is channeled to vented floor tiles in the cold aisles of your DC. Hot air goes out the back and with the right airflow rises to get sucked into the HVACs, cooled and recycled back to the fronts of the racks via the floor vents in the cold aisles.

Within each rack you need to be careful to not leave gaps inbetween equipment without blanking plates because that will fuck with your airflow.

This is the old school way of doing things. More info here: http://www.cisco.com/c/en/us/soluti.../unified-computing/white_paper_c11-680202.pdf

This is one variation of the new way of doing things (using curtains or sliding doors)

cold-air-plenum.jpg


There are loads of ways to do this, some better than others.

I will never permit water cooled racks in my DCs. I've never heard of a problem with them before, they work very well. I've just had too many floods from other sources to risk adding another one :D

Other than airflow, n+x redundant cooling units are essential. Air lock doors help in big DCs.

Common oversights are forgetting to factor in heat given off by lighting and also engineers working in the room. A human at rest gives off 100-200w of heat. This is a bit more if they are working hard loading up SAN chassis and collapsing carboard boxes!
 
I didn't answer your question. Yes, I think ground water cooling and reverse heat exchangers could be used to lower power bills but not as primary cooling mechanisms.

Don't get me started on ideal humidity and temperatures. That is a topic for considerable debate!
 
Stonehedge, you certainly know your stuff, I can remember doing a calculation in University for heat loss through walls, and basically almost all of your heat is lost through conduction of the surfaces, of course if the walls are well insulated this will exacerbate the problem.And the walls will only transfer so much depending on the heat differences and the wall composition.Am surprised 200 watts of body heat makes much difference as you gotta be talking many thousands of watts in these large server setups.
 
Thanks!

Ideally, the air in your room won't be heating the walls up so you shouldn't have that problem....just a big electricity bill! What you say is correct though. I had to replace and then decom a large and periodically failing data centre built in a brick building dating back to the very early 1800s or late 1700s in 2009. Lets just say they built walls thick back then! We had cooling problems and whenever the air conditioning failed, it'd take up to 48 hours for the walls to cool back to ideal ambient temperature. I wish I could tell you more about that project but sadly I can't.

You're right, lighting and person heat output is tiny but considering that both can also contribute to turbulence, I always factor those loads in too.
 
Which is why the future has to be using our mobile phone's, 50% of the CPU & Battery lets say to help secure the network...
 
Who on earth has managed to rack up 10 million invalid shares on the official pool? I managed to submit 4 million invalid shares in under a minute last night but I got booted because of it! yibble?
 
Hey Stonehedge,

I've located a few suspects. However, the Stratum server doesn't log everything. Or more accurately, to have the Stratum server log everything would cause a degradation in service.
 
Back
Top