Why Keep a server room cold?

TheBluePill

2[H]4U
Joined
Aug 17, 2005
Messages
3,773
Here is a question that has been put to me and I belive I have the answer.. but i thought I would ask some of my fellow OCPers about it..

I have a server room with 5 Poweredge servers, several large UPSs, 10 Catalyst Switches and an HP K Class 9000 server..

I Keep the thermostat set at 60 degrees and the room stays fairly cold.

Management wants to "cut costs" and wants to know why we can't set the temprature to 70 or 75 degrees in the server room. (which is madening to me).. But they don't want an "If" or "Could Extend drive life." etc.. they want a hard reason why I want the server room to be as cold as possible..

So far, my research tells me that the Operating Temps of my PE Servers is 50F to 95F and my HP is 68f to 86f.. I also have 1400w worth of Power supplies in each PE and a fairly large requirement for the HP. That is not counting all of my Cisco Gear putting off heat in my network cabinets..

So mathmatically, how can i give them some hard data to confirm just what temprature I should have my server room, with a reasonable margin for safety?
 
Without any maths I would ask them to keep it at 65. That will knock a third off the AC bill. You could also try to improve ventilation in the racks
 
Hard data? Let them switch the AC off and have them see the results.


The only hard reason is going to be hard to come by, as it depends on many variables and will depend on a large historical data of failures of drives and components.


An argument you could give them is that the room is sealed, thus regardless of the temperature, to keep it consistent would be very close to the same cost. The only heat produced is from the walls or the servers, and since the walls would be the same amount of radiation and so would be the servers, that to keep the level down low would be in essence the same? I don't know, just try and swing them something like that. All you are doing is keeping the threshold low, your loss of energy is really low since it's a smaller room and most likely well insulated and sealed.

You could always convince them of new more efficient servers?

Another thing you can do (don't do this) is to make them aware that the real cost cutting is in the upper management salaries. he he.
 
Thats what I am afraid of.. This is my area of expertise, but we have a guy with a PhD in wood science that sticks his nose in anything that he thinks will make him look "smart".. Every now and then he cooks up some idea that he makes a big fuss over that is really nothing.. BUT.. I do have to come up with a reply..

So, I can try to explain how keeping drives cooler extends their life, but is there any article or study on the matter?
 
Hard data? Let them switch the AC off and have them see the results.

That's exactly what I would do...seriously. Show them where all the important temps are now, shut it off and make a comparison....
 
The problem is that if you give in, next month they are going to see some savings and say "hey, why don't you make it 75!?"
 
Explain to them that processors have heatsinks and cases have fans for a reason. If the heat can't be dumped outside of the room, it has to be removed in another fashion. Cut a deal to crank it up to 68F if they agree to chop a hole in the wall and construct a direct-vent system. ;)
 
Get some yogurt. Put it the in fridge. Then remove it from the fridge and let it sit in the summer heat for a few hours, and then tell them that's what happens to the servers as well. :)

Overexagerrated(?) perhaps, but I think it gets the point across to the layman. Better if you pick some food that could be analogous to a critical component. "Oh look, your lunch spoiled. Now imagine that our data went to hell too!"
 
Thank you guys. I have collected a bit more data;

I used a raytek Minitemp meter to see how hot things were running and take some temps around the room.

The Room:

AC Vent Air Temprature is 62f
Floor Temp: 69f
Ceiling Temp: 73f

HP 9000 Server Temp (Inside Case): 76f
Dell PE Server Temps (Inside Case): 78.5-82f
Cisco Catalys Temps (External): 86.5f

So we are within 10 Degrees of maximum operating temprature of both our Dell and our HP servers.
 
Where I used to work our server room had no dedicated AC for awhile. Even on new servers I had drive failures in droves. Seriously, I'd lose one or two drives in a RAID array every six weeks.

You know what it's like to lose two drives in a RAID 5 array at the same time or nearly the same time? It sucks. It sucks even more when you lose one drive and you order another, (we didn't keep them on hand.) and then while the replacement is in transit, loosing a second drive and killing the entire array. (5 drives in the array to start.) It's funny, they put an AC unit in after they lose a ton of data on multiple occasions and each of these drives they were killing were $600 a whack. They decided it might be a bit warm in there. Sure enough we put the AC unit in and we'd go a year or more without losing a drive.
 
score one for the cooter, that link looks like it's loaded with B.S they'll never understand and will agree with. I'd run some numbers by them to.. A new processor is X ammount. and data lost. Downtime, maintence cost.. all little fees here and there but would prolly equal out to cost the same if not more to just run the room at 65*... then you have time wasted on downed components;
 
If they force you to up the thermostat just occasionally shutdown a server or two and act like the heat is causing it.
 
If they force you to up the thermostat just occasionally shutdown a server or two and act like the heat is causing it.

Just walk behind it and yank the power once in awhile. :D
 
Do you use a company for your disaster recovery plan? Like Sungard or Iron Mountain. If so contact someone there, they will be more than happy to hook up up with the resources you need. If not, Try disaster recovery journal for articles on the matter.

When i was in a data center i kept it at 72 with 30% humidity.
 
THANK YOU ALL for the help on this.

I took some data from some of those links and all of my temprature measurements.. Management has decided that turning the thermostat down to save $100 a year just isn't worth the risk.

Thanks
 
If they force you to up the thermostat just occasionally shutdown a server or two and act like the heat is causing it.

worked in my situation. they refused to leave the ac on overnite here.

id shut down systems remotely, id disable network interfaces...etc...'pretend data loss' then 'recovery' the bosses were pleased, and finally kept the ac on. no more downed systems, no moe dead disks...no more commin into work at 8am and the temp is 85+ either :rolleyes:
 
worked in my situation. they refused to leave the ac on overnite here.

id shut down systems remotely, id disable network interfaces...etc...'pretend data loss' then 'recovery' the bosses were pleased, and finally kept the ac on. no more downed systems, no moe dead disks...no more commin into work at 8am and the temp is 85+ either :rolleyes:

Pwned! Mess not with those who control your data, for he who controls the spice, controls the......oops, went on a Dune binge. Seriously though, nice.
 
i saw an odd bit of info.

google experianced more hard drive failures in colder temps than in hotter temps.

wasn't a scientific study or anything, but it made google go "wtf?".

edit:
found a reference to it, so yall know im not pulling it out of my ass.
http://www.engadget.com/2007/02/18/massive-google-hard-drive-survey-turns-up-very-interesting-thing/

and the pdf
http://labs.google.com/papers/disk_failures.pdf

hey and the pdf has some info that the op is looking for.

I'd agree that the drive failures will show correlation between manufacturer, model and age more often than anything. For example any of the Maxtor drives that are built like Quantun drives are shit. Period. (As an example.) As far as the cooling thing goes, what that says and what my experience has shown me are very different. I think that if we are talking about slightly warm and slightly cool drives you may not see a difference and excessive cold or excessive heat would destroy drives quickly as well. I'd have to say that in my expereience a drive with some simple air cooling on it is better off than a drive with no cooling.
 
those hard drives you're worried about have factory suggested operating temperatures. if the temperature of the drives does not reach critical levels if you up the room temperature from 60 to 70 *F and the drives still remain in the safe operating temperature range, than you have nothing to worry about.
 
I think he is more worried about the extra work needed when the servers fail becuse of the heat. :p:D
 
worked in my situation. they refused to leave the ac on overnite here.

id shut down systems remotely, id disable network interfaces...etc...'pretend data loss' then 'recovery' the bosses were pleased, and finally kept the ac on. no more downed systems, no moe dead disks...no more commin into work at 8am and the temp is 85+ either :rolleyes:

Please tell me you quoted them 4 hours then got it working in 1? You know, the Scotty miracle worker thing?
 
I prefer that the average temperature of our cabinet inlets is around 65 degrees. Our A/C units put out about 54 degree air. I have racks and racks of equipment though and over 30 tons of cooling. Our IBM servers along with our Hitachi SANS put out the most heat. There are far better ways to cut costs rather than screwing with the A/C. However I work at a larger company and just spent $31k on additional cooling this past week.

Here are some reasons to keep the room cooler.

1. The lower ambient temperature will give you a longer warm up time should something happen to your A/C unit(s).
2. The coolness of the room does not equal the coolness of the cabinet internals.
3. In the event of a power failure, and you run on UPS - many times your A/C will not work. Keeping the temperature lower again buys you some additional time.
4. Drive life IS prolonged in most cases when you keep them as cool as possible.
5. Determine how much equipment you have and it's value, along with the services provided. Then figure out how much downtime and equipment replacement costs would be if you were to lose cooling. Hopefully you redundant cooling of some sort. Does the cost savings of changing the room temperature provide more IRR than if you had to replace some equipment?

Personally I have a large diesel generator, large Liebert UPS, and all of my A/C units run on the generator in the event we lose power. That means in a power outage, I only lose cooling for about 5 seconds while the generator fires up and the transfer switch takes us off of commercial power. I cool all of our IDFs with our Cisco gear as well with smaller Mini-Split Bryant systems, keeping those rooms at about 70 degrees. I just can't see saving a few bucks versus having to buy a replacement supervisor for a switch, or having a server just up and die. When the temps hit 90, BAD things start to happen. Don't ask me how I know this.

I started off with a small Data Center, and once you have one over heat, you will fight the fight to keep things cool. I don't over do it though - I set our thermostats at 72 degrees within the data center, keeping in mind that they are on the walls and not where the supplies.
 
Well guys, there is a big difference between turning OFF an AC system and simply raising the temps slightly... There is no need IMO to keep a server room at 60 degrees. Heck, I've been in several data centers with thousands of computers and none of them were THAT cold. 75 is too warm, but something like 68F is very reasonable and is plenty cool enough for an equipment room.
 
I've been in one renderfarm server room for a famous company, and the temperature felt like a walk-in freezer from the back of a good supermarket. And with the racket the blade servers were making (thousands of them!), it felt like I was standing in the machine room in that supermarket, at the same time.
 
The problem is you don't keep a server room at 60 degrees, unless the 'server' room is tiny. As server load varies, so do the temperatures they output. You have some racks putting out more BTUS than others. Cooling a computer room is a bit tricky, but a 60 degree cabinet inlet temperature does NOT equate a room being 60 degrees. Parts of the room could be 80, other parts could be 55. It depends where you're at, cause the cabinets take cold air in and blow warm air out. For example we dump all cooling on the inlets of our cabinets - we do not cool the exhaust sides. So if you walk down behind a row of cabinets, it's a bit warm. Walk down the inlet side and you're to look for a jacket. If your thermostat is close to the A/C supplies (cold side), then I'd say a 67 degree temperature setting would suffice. If it's closer to the returns (warm side), then I'd say a 72 degree setting would be a good starting point. I also have our units configured to run the fan 24x7 even if the compressor is not running generating cooling. That way there is still some air circulation, otherwise the hot air will pool in one area behind a rack of cabinets and not have anywhere to go.
 
Back
Top