2

My company has around 20 servers which we want to operate in an underground room, where the temperature over whole year ranges from 5°C to 28°C.

The only thing I am worried about is that in the summertime, the place will be relatively humid, around 80% RH. Will this cause a big impact on the stability of 7x24 running hardware OR shorten the life span of the servers?

John Gardeniers
  • 27,262
  • 12
  • 53
  • 108
c2h2
  • 759
  • 2
  • 8
  • 20

2 Answers2

12

Check the server specs. They should (or at least big OEMs like Dell, HP and IBM do) come with operating ranges for temperature and humidity.

Three things though:

  1. Putting servers in a room that's not climate controlled seems like a bad idea in general, and I wouldn't do it. This specific room sounds incredibly bad, like it may as well be a kill room for electronics.

  2. At 5 degrees Celsius and 80% humidity, you will run into issues with condensation occurring, and water condensing on your electronics will cause hardware damage.

  3. At 28 degrees Celsius and 80% humidity, this specific room sounds like it is probably going to overheat your gear, which will either damage it, or cause it to shutdown to prevent damage. The room will get hotter as a result of having electronics running in it, and without a way to dissipate the extra heat the gear produces, you're essentially creating the same effect as an oven.

High humidity will probably compound heat issues, because of the greater energy it takes to heat or cool humid air - it will take longer to heat up initially, but once it heats up, it's going to be much more energy or time-intensive to cool down.

Regarding the temperature range discussion brought up in the comments, there's an older thread on it, here. And it's all well and good to say "Google does [blah]," but Google also runs an expected failure model on cheap, commodity gear with many millions of dollars invested in their custom software to make it work. What Google does is not applicable to the vast majority of situations. (Such as asking about running 20 servers in a non-climate controlled room that wasn't designed with servers in mind. For example.)

HopelessN00b
  • 53,385
  • 32
  • 133
  • 208
  • I would +1, but seriously - 28 degree are a joke. Any decent server has been rated for more than 30 degree because it can operate in a normal office. In summer. And modern servers are rated for 45 degree. Celsius. So they can operate in a higher than ambient temperature in summer in most areas, which seriously cuts down cooling costs. Check your facts. I can not ever remember a machine that had problems with 28 degree. So, no +1. – TomTom Jan 08 '13 at 08:15
  • 7
    @TomTom I think he's talking about ambient temperature and not internal temperature. – Lucas Kauffman Jan 08 '13 at 08:17
  • 5
    @TomTom I just don't even know what to say to someone claiming that an ambient room temperature of 28 degrees is fine for a server room. Uh, "good luck with that," I guess. Must be handy when you want to cook food and the lines for the microwave in the breakroom are too long. :/ – HopelessN00b Jan 08 '13 at 08:18
  • 2
    @HopelessN00b PLENTY of research was done in this area the last years - to not COOL a server room but just replace the air. Building data centers this way is standard these days - you never pump the air out, you just replace it with fresh air from outside. And outside can be hotter. Check http://www.supermicro.nl/products/nfo/FatTwin.cfm for some SuperMicro servers. Ambient Temperature per spec: 0 degree tp 47 degree (!). It is called "Free Air Cooling". SERIOUSLY saves money and is standard now. Intel and Google did big scale studies the last years. – TomTom Jan 08 '13 at 08:21
  • @LucasKauffman So what is wrong with the envelope from your servers? I mean, my OLD one are fine with more han 30 degree as per their manufacturer, and my new ones run up to 47 as per manufacturer. That is ambient. – TomTom Jan 08 '13 at 08:24
  • always a good read: http://www.datacenterknowledge.com/archives/2008/10/14/google-raise-your-data-center-temperature/ – TomTom Jan 08 '13 at 08:27
  • 2
    @TomTom Well, if you completely ignore the fact that 28 degrees is the ambient temperature of the room before moving 20 servers into it, you have a point. But that's not the situation asked about. The situation being asked about is whether or not a room with an empty ambient temperature of 28 degrees will work. And it will not, especially considering the other factors, like it being underground, which makes it bad at dissipating heat, without a cooling system thta's very clearly not present here. – HopelessN00b Jan 08 '13 at 08:52
  • Actually, as you can see in my answer, the ambient temperature is totally irrelevant - it could be 0 degree. What is needed is to know where the balance point will be. I would strongly suggest that 29 degree or degree wont make a difference -the 20 servers ANYWAY overload the room. – TomTom Jan 08 '13 at 09:01
  • @TomTom but he's not even talking about displacing air, it's just a room also we don't know the volume of the room. I don't think we can asses if it's good or not. We cannot tell from the current given question if there is some kind of ventilation in the room possible to at least extract hot air. – Lucas Kauffman Jan 08 '13 at 09:05
  • 2
    +1 but it's worth noting that commercial grade electronic chips (that's the lowest standard grade there is) have an ambient temperature range *during operation* of 5 degrees C to 125 degrees C. However, some other components may be limited to as low as 80 degrees C. Bear in mind though that this is referring to the ambient temperature *at the component*, which will be of course affected by the heat generated by other components in the system. – John Gardeniers Jan 08 '13 at 09:44
5

Ok, here we go. I would absolutely not be concerned with temperature IF that is allowed for the machines, same with humidity. Modern servers support up to 47 degree ambient temperature and even old ones are made from components that are mostly also used in tower servers and allow ambients above 30. Intel and Google studies show no bad things happening.

Most important is to check what he manufacturers allow.

The problem is secondary stuff, MOSTLY USV - Batteries may not like this temperature.

The other problem you may have is delusion about what temperature IS. The underground room may have a low temperature because it is shielded from sun etc., but that may not give you a VERY big amount of dissipation - so you may simply heat that up VERY fast the moment you actually emit heat in it. My own server room is in the basement of our offices - well, 1 meter over ground, 1.5 meters below. NICE cool room. 4 servers and it heats up. Why? Because it is mostly cool because it is isolated and the ground is cool, but at the same time the heat does not dissipate in hugh amounts. For 20 servers that LIKELY does not work - you will overheat the room without active cooling component, which costs have to be factored in.

In our case we are installing air conditioning now and will ramp the excess heat into the house heating system ;) We also removed all batteries from the room (external large USV installed) and will increase ambient temperatures in the server room to the 40 degree level as we replace servers. That is still way below what the manufacturer or our hardware allows (47 ambient) but it will make active cooling QUITE efficient - we never had that temperature outside, and cooling with a cooler outer side is more efficient.

If you NEED active cooling, then humidity is a non-issue anyway as air conditioning can remove humidity.

So, the main question is - how much heating do you create, and that is directly the power consumption of your servers. Assuming 200 watt per server you run about 4kw. That likely is above your dissipation capacity.

So, the whole thing is not as easy to answer. Anyone focusing on temperature here so far totally forgets that temperature is not absolute, it is a balance between emitted heat and dissipation to environment. And the current temperature may have no meaning the moment you add the servers. It is unlikely you can add 4kw energy to that room without getting into active cooling of some sort (and air replacement is active cooling unless you manage to create the airflow just by heat differential - possible, but that requires building the building in a very special way).

TomTom
  • 50,857
  • 7
  • 52
  • 134
  • +1 for airco can remove humidity – Lucas Kauffman Jan 08 '13 at 09:08
  • thanks for your answer, I have actively measuring the total power consumption of my servers using wattage meter. when idle, average power per machine is around 50w, and peak 200w, therefore around 1kw~2kw in total. In summer, I hope the total generated heat will drive down the RH%. will monitor that, and install air con if necessary. – c2h2 Jan 09 '13 at 06:17
  • Ok, that is VERY low - sorry to say, but it smells like you should look mmid term into virtualization. i can not imagine having servers with that low power - my typical servr has 1-2 processors, 64+gb memory OR is a HPC node. Reason is - it is a lot cheaper and easier to manage to have fewer servers. But still - 1-2 kw MAY stress that, but you have to find out. Airco may at least get rid of your humidity ;) – TomTom Jan 09 '13 at 06:42