Slashdot Mirror
What's the Proper Temperature for a Server Room?
Izzard asks: "As a network engineer, I sometimes have to spend many hours in other people's server rooms. One in particular has a good few servers, DVD jukebox, plenty of monitors and switches etc. It's a thick, stone-walled room with a big door. It would get very, very warm were it not for the two huge air conditioning units in there. Someone has decided that these units will be set to maintain a constant air temperature of 17-18 deg. C (62-64F). After an hour or so of sitting in the air stream from these units my legs go numb and fall off and I can't type. Now my guess is that it would be fine to set the units to maintain, say, 21-22 deg. C (70-71F) to make it more comfortable for those of us who have to work in there. This argument comes up a lot, and my position is that the room doesn't need to be refrigerated, *per se*...it only needs to be prevented from overheating. Consequently I maintain that a *consistent* temperature of 'pleasant' for the room is almost as good a consistent temperature of 'a bit nippy'. Is there a definitive answer to this?"
...at the radio station I work at, the main studio containes a bunch of sensitive electronic equipment like OnAir Consoles, computers, standalone editing stations... and that room is kept 24-7 year round at a constant 71*F. Not too cold to stay in for extended periods of time, and not too hard on the AC unit, but keeps the equipment running in great condition.
After an hour or so of sitting in the air stream from these units my legs go numb and fall off and I can't type.
You type with your legs that is awesome.
42.
Why not fork?
...stay out of the data center. While I can't backup with studies, or any hard scientific evidence, I can support that with experience. A real server room will have hot spots, sometimes 20+ degrees over the mean temperature the room (especially poorly ventilated ones. Big ACs don't equal good ventilation, btw.)
These hot spots can (and are often) murder for server farms. Take a page from the experts (i.e. big colo firms): keep your data center cold, and have lots of airflow near the racks.
This isn't necessarily always true, and a small data center can probably afford to not be frigid. But if you've got a lot of money in your data center which would you prefer: big, expenive AC bills or more expensive outages on very expensive hardware?
4.2 kelvin. liquid helium. turbo charge your servers with superconductivity.
Everyone is born right-handed; only the greatest overcome it
Otherwise, keep it at 65 degrees, and either learn to deal with it, or learn to bring a sweater with you.
Hey, c'mon now! Give him some advice he can actually use! He specifically said "After an hour or so of sitting in the air stream from these units my legs go numb and fall off and I can't type." You should be advising him to "learn to wear pants" instead!
GMD
watch this
"Is there a definitive answer to this?"
Buddy... this is SLASHDOT.
At least datacom equipements. PC and big irons may be a bit more flimsy. But, for what I know of a BIG vendor of datacom gear, all the products are specified to run from 5C to 40C and are actually tested at -5C (23F) and 55C (131F) ambient for weeks in a row. The nominal operating temperature is assumed to be 25C.
;-)
So from 17C to 40C, there's quite some room. Yet, watch out. Those temperature are specified at the cooling inlet of the equipment. With lateral cooling, the gear at the wrong end of a row of racks may suck heated air from the other racks and see much higher temperature than ambient. That's bad room design
IMHO, the guy who spec'ed 17C is overdoing it (and padding the wallet of the local utility). 25C should be OK.
SNS Not Sig
I've never understood the need for super-cold machine rooms. 70F is well within the healthy operating parameters of most computers.
Let's also not forget that air conditioners aren't always good for the mechine: They increase the level of static electricity in the area, and can blows dust around if don't have an airfilter in that room. Dust and static buildup can harm the machines more then at 70F temperature.
As another poster pointed out, the only good reason to keep an AC at a really low temperature is to deal with the hotspots. It may be 60F at your desk, but the big server is sitting in a corner far from the AC, and it's case is 80F. But that's just a result of poor air flow, and can probably be resolved with a few fans and a few air ducts.
"Can of worms? The can is open... the worms are everywhere."
I pay close to $120 a month to keep my apartment 62F year round, and I save a huge amount of money because I don't have to constantly repair and replace equipment.
Really? I too keep my apartment at 62 F year round, and I've been having problems with some of my more sensitive equipment. In particular, I'm noticing some shrinkage. Do you have any advice?
-
The economics include the initial cost of the system, the
continuing cost of energy & maintenance, and the cost of
failure - all of the servers overheating and the network
going down.
-
The estimates include the heat load (heat energy output of the equipment),
the system effieciency over time, system neglect, & management cheapness.
-
The engineering includes doing all the math to determine many of the
previous factors, & compensating for all of those unknowns with a factor of safety.
:)
Is there a definitive answer to this?The answer to your question(s): Get an HVAC engineer to look the system over, and tell you if the existing AC is overkill. (You might want to be sure that she doesn't think that she'll get the contract for any additional work needed. :)
'Course, this might all be quite accademic. The reason for the chill might be that your boss's brother has the contract, or that your boss had a bad experience with a server catching on fire. Then it's not engineering, it's psychology. :)
"_I_ spend $1400 on A/C every year. Aren't I cool???"
As a matter of fact, yes.
Moderate drunk! It's more fun that way!
Most of the labs I've built or worked in, we've set up the ACs somewhere around 70-72 degrees. It's plenty cold while still leaving a small amount of headroom. The headroom is useful in case there's a sudden influx of heat and it takes a while to restabilize the temperature. Can't really predict what sorts of wierd things might cause that - a crapload of new equipment, an A/C unit going out of service for an hour, etc, etc..
It's very important to get your humidity correct as well. If the humidity is too low, static buildup becomes easier and static damage more frequent. If it gets too high, corrosion occurs faster. Computers like to be in the middle, if I remember right the ideal for most machines is around 35% humidity? It's been a while, that might be off by a bit.
Don't forget the whole BTUs thing. All your equipment will have a sticker or manual (or call the company) saying how many BTU of heat it puts out at max. Add them all up and make sure you have enough A/C capacity to account for the BTUs during a failure scenario (e.g. Buy 3 AC units that can handle 1/2 the desired BTU, so you have N+1 redundancy). Be sure to estimate the future as best you can in the BTU calculations - replacing A/C units when upgrading new servers in a fully loaded production room can be a bitch.
Computers like stability too, so try to set it up such that the humidity and temperature stay constant while all your gear is running. If they're wobbling up and down throughout the day or week, you need to fix it. You can buy cheap chart recorders for this, they drag a pen over a graph and show you a temperature line for a week or more. Assuming your A/C is adequate for the BTUs, the wobbling is most likely from bad airflow design.
Airflow design can be a black art, so you might want to get a professional. In general, most datacenter-class machines suck cold air from the bottom and/or front and exhaust out the top and/or back. Space out your vent tiles, too many too clsoe together can shunt air away from the inlets on your equipment. But by all means place vent tiles here and there in the empty areas to even out the room.
And if you're looking for professionals to do these kinds of things, up to and including designing and building new datacenters from scratch, I can't recommend IBM Global Services high enough. They really kick ass at these things. It almost makes up for AIX sucking so bad
11*43+456^2
I used to work as a HVAC tech at Wright-Patterson AFB in Ohio. As such a good portion of the equipment we maintained is for computer rooms (including a Cray system).
/kc
From the HVAC viewpoint, two important factors are the heat load vs. capacity and the design of cooling equipment.
Air conditioning units loose compressors frequently due to "slugging". Slugging is what happens when the refrigerant is not fully boiled off as it leaves the evaporator and returns to the compressor. If liquid reaches the cylinder of the compressor, the reed valves get bent (in a reciprocating compressor) rendering it useless. New "scroll" compressors can deal with this a bit better however, they too will fail in some circumstances. Simply put, the warmer the room, the more likely that all the refrigerant is boiled off before it reaches the compressor and the accumulator. With the room chilled to colder temps, the refrigerant may not boil off fully due to the small temperature differential between return air temp and discharge air temp. chances go up in the potential problems that can occur. This directly relates to the life expectancy of the compressor. While high quality air conditioning units are designed against such conditions with accumulators and suction pressure regulators, a regular air conditioner typically has no "anti-slugging" devices because they are designed to run under very specific conditions (outdoor 70-100+ and 72-82 F indoor temps. (something to keep in mind if you are thinking installing a run of the mill residential/commercial air conditioner unit in your server room to save money).
BTW, over sizing the equipment can be just as bad as under sizing. Oversize the equipment and it will short-cycle rapidly thus wearing out the mechanical and electrical components, humidity control issues, and slugging.