Slashdot Mirror
How to Run a Computer in a Sub-Zero Environment?
Underdog asks: "I've seen tons of Slashdot articles on cooling hardware, but my company may be taking on the task of wiring a large sub-zero (as low as -14) warehouse with temperature sensors and the requisite network equipment and computers to read them. Our initial proposal includes at least a dozen acquisition computers, hung from the racks in the freezer. Does anyone have any experience with installing computers in extremely low temperature locations?"
I once worked for a company that had a computer closet on top of a mountain.It would often get -25 to -50F and sometimes much much lower. If you can find a way to enclose the computers they will keep themselves warm. We just put up some 1"-thick insulation inside the walls of the little shed and the two computers kept it at 40-50F in there at the worst times.
-----------------------------------------
Remove the Greed which plagues mankind.
This may be missing the point of the OP, but why not install the computers elsewhere and use something like the Sensatronics sensors? The sensor device can be outside the freezer - only the probes need to be in that brutal environment. The device connnects via Ethernet. We monitor using Intellipool Network Monitor, although before we had that package I threw together a Perl script to poll the devices via snmp.
The hard disk is the big problem. It will produce enough heat to keep itself warm and working if installed in an insulated box. It must be kept on at all times or an auxiliary heat-source like a light bulb must be provided when the drive is powered down.
Mea navis aericumbens anguillis abundat
Military equipment often uses conformal coating, which is a spray-on plastic coating that protects the components from the environment.
Mea navis aericumbens anguillis abundat
When some overclockers use sub-zero equipment, condensation becomes a big issue. With stuff like LN2, some OC'ers dump the entire motherboard into a non-conductive tray of oil. You could look into something like that.
"Sure there's porn and piracy on the Web but there's probably a downside too."
Transistors are designed to behave within a specific range of voltages and switching speeds for a particular range of temperatures. Most COTS electronics are targeted for an ambient temperature around 72F and work best at that temperature. When temperature extremes are needed, the transistors are actually doped and constructed differently.
That having been said, there are some things you can consider:
1. Do the computers really need to be in the freezer? If there is a way to build it so that they're not in the freezer, do it.
2. Enclose the cases with no ventilation. At subzero ambient temperatures they'll lose enough heat through the chassis. Insulate until the internal temperature is reasonable but not so far that it'll retain too much heat.
3. Install an electric heating coil in the case to bring the temperature up if it drops too low.
4. Underclock everything on the system: the CPU, the PCI bus, etc. Stretching out the clock cycles should give you a greater tolerance to the change in how the transistors behave and lower than expected temperatures.
5. Don't forget to consider the impact of the heat load on the freezer. You said computers with an S. Each one is going to dump 200 watts or more of electric heat into the freezer 24/7. Does the freezer have enough excess capacity to handle that and still do its job?
Moderating "-1, Disagree" is simple censorship. Have the guts to post your opinion.
Putting heaters (computers) in an environment meant to be cold is just adding to the cooling workload. If the computer is at any decent operating temperature, it's going to be heating up the immediate surrounding area, and you don't want that.
Put the computers outside. String sensors as needed. If you have to have electronics near the business end of the sensors, put those electronics under the floor or over the ceiling.
Think of your refrigerator. Would you put even a small computer in there to keep your food warm?
Infuriate left and right
Correct. Condensation will only form on objects which are below the dew point. Someone more qualified than myself could give a better explaination, but the dew point (usually as a temperature) is defined in terms of the barametric pressure and the humidity. Also, at very low temeratures (ie below zero), you really don't have to worry about the amount of moisture in the air. The partial pressure of water vapor at freezing is 0.61kPa (6.1mbar). As a comparison, one atmmosphere (1atm) is 101.3kPa, and the partial pressure of water vapor at room temperature (usually 22 or 25 degrees C) is about 2.9kPa.
I ran across this googling "industrial pc for low temperature environment" (without the quotes):
It's specs say it has an option to go down to -20C operating temperature.
Where is this magical water you're talking about coming from? The air in the computer is air from the environment heated up. There is no source for water inside the computer, and it's warmer than ambient. If there is condensation it will be outside the computer.
That which is done from love exists beyond good and evil
For what it's worth, we've always built room-temperature enclosures to house electronics gear and PCs for the work we do in Antarctica. It's almost always easier and cheaper then trying to insure all your equipment can survive harsh temperatures.
For the odd piece of gear that needs to survive out in the open, we test them thoroughly in a freezer ahead of time. Some things - in particular simple solid state single board gear with no moving parts - seem to do quite well down to -50 C or lower. But, as capacitor values drift and sockets and connectors contract, even some likely candidates fail. Anything with lubricants or precision mechanical parts (drives, fans, etc) are almost certain to cause trouble. Expect your batteries to die and a some read-write storage media to fail.
But, is it really necessary to put a dozen full computers in this environment? It sounds like serious overkill to run a bunch of temperature sensors. If you absolutely need to use PCs, see if you can place them just outside of the cold space and run cables. Or, if that's not possible, put them all in a single, insulated, enclosed space with an active thermostat and some electric heaters. Make sure that when all the PCs are running at full tilt the temperature in the box is slightly below your target, so that you can control it with only a heater.
Better yet, replace the PCs with small readout and control boards. If all you need is to record temperatures to within a few tenths of a degree, building a board that will give you dozens of channels and a straightforward digital interface should be a few day's work for a reasonably competent engineer - and fabbing them may well cost less than a dozen PCs. You can then hand pick parts and packaging that is rated (or tested by you) to low temperatures, or you can build in very small heaters that keep individual parts warm without dumping too much heat into the environment. You may even be able to find such a product off-the-shelf if you hunt around.
If you absolutely must have PCs, see if you can't find a small single-board computer that will do the job. Test several over dozens of thermal cycles in a freezer before deciding to use it, and buy a bunch of extras.
If I was riging it up, I'd use something like RS485 into sealed units with a small custom board in a sealed unit with the sensors hanging off of that. Then you only need one or two PC's outside plugged into the networks of sensors to read off the data and log it.
p ?model_id=1-1TWNLI
& ArticleID=6191
You could plug pretty much any PC with a serial port in, with a converter like:
http://www.advantech.com/products/Model_Detail.as
The only dificulty left is working out what kind of connectors you can use, if it's all hard wired, then it might be fine to wire the cables though sealed gromits into the boxes for termination.
The protocol could be quite trivial too, say send a couple of characters like R521,53 to say you want to read sensor 53 on unit 521, it'd run out over the bus, get picked up by the right unit, and reply a short time later with something shocking like V521,53,258 (where 258 is -15 degrees in kelvin).
But don't take my word for it, just build a low temprature version of:
http://www.elecdesign.com/Articles/Index.cfm?AD=1
Computers are hurt by cold. If the operating temperature is too cold the chips and other electronics will heat up and expand faster then the pcb causing solder joints to break. Google for thermal expansion.
It's not just water that you have to worry about. Commercial-spec solid-state electronic components are typically rated for operation between 0 and 70 degrees Celsius. Electronic components conduct electricity better (or worse, in the case of many semiconductors) at lower temperatures, so even in a humidity-controlled environment, you could end up melting certain components.
What you need is either computers that are built entirely out of industrial or automotive-spec components that are rated at -40 to 85 degrees Celsius, or you need a temperature-controlled server room that will keep the computers within the commercial-spec range. Both are going to cost money.
http://outcampaign.org/
I've done stuff like this with computers for balloon payloads that go up in the stratosphere, where it's around -50 C. Here's two tricks that should help. Trick number one is build a box out of styrofoam building insulation and duct tape. Assuming you're in the US, you'll see a number printed on the insulation like "R5" or "R10." That's the thermal resistance, in BTUs, hours, deg F, and square feet. No, I'm not making that up. Guesstimate the power dissipation of the computer and use that to make the first design, then test and iterate. You'll want to stick a thermometer on the case or other convenient location. If this isn't reliable enough, then trick number two is design your insulated box to run a bit cold, and build a thermostatically controlled heater. We usually designed our own, because we like to do things the hard way, but I believe at someplace like Newark Electronics you can buy a little package that contains a heater and a bimetallic thermostat, you just supply the power.
I've done lots of industrial installations, in places where -14 would be considered 'toasty warm' compared to outside temps in the middle of winter. If I saw a proposal that includes putting full blown computers in the freezer, the first thing I'd do, find another vendor, this one obviously has no clue when it comes to embedded industrial equipment. Mil grade sensors that are good to -40, may not be a dime a dozen, but, there's lots of them out there that you can just buy and install, which will happily feed the data back to a computer sitting in an office somewhere.
The bottom line, if you are going to put rack mount pc's inside the freezer, do your customer a huge favour, and reccommend they find an expert in the field. You will be saving yourself a long term support nightmare, and your customer a whole big pile of money, because the proposed solution is kind of like taking money and flushing it down the toilet.
Most computers should work fine by sustaining themselves with their own heat, but I wouldn't power up a hard drive that I cared about if it was below freezing. I would try to find a tiered power-up system like hard core liquid cooled system use. These go between the power switch and the motherboard, so that powering on first pre-warms the components, and only when they got to an acceptable temperature does the system power on.
Wonder if peltier pumps would be handy since you can simply reverse the current to reverse the heat/cool direction...
I had the pleasure of setting up a couple of systems in an unheated office in Maryland, in Februrary. It was "only" 20 in the room, but the hard drives did not want to spin up until they warmed up. Aparently there is some sort of lubricant on the platters that turns to GLUE at 20 degrees. So.... Put the servers outside the cold area. Make everything in the cold area diskless. My father used to work for a company that made cockpit voice recorders. The bay the recorders get mounted in is unheated and unpressurized, so it gets 30 below and very low pressure. To compensate for that (and for condensation) they burried the entire circut board in a block of epoxy. If you run terminal server, you may be able to find a dedicated terminal server client that you can just bury in epoxy. The hardest part will be the monitor. As far as I know LCD will freeze at that temp and just not work. CRT will potentially have condensation problems. I don't know enough about how plasma works to know what that will do.
Of course, if the air is guaranteed dry, then it's another situation entirely. Dry air can be any temperature above the minimum functioning temperature of the components, and everything'll be just fine. A warehouse keeping things that could get damaged from condensation or ice, for example, is going to be extremely careful to keep the moisture out. If that's the case, you don't need to repeat the process. Let the computer chill out.
Immersion methods won't work well if you have mechanical components, such as hard drives. You also have a major problem of the bearings freezing up. So, if these are "traditional" PCs with mechanical devices, you have to go for a different approach. In these sorts of cases, you really want to have the computer lagged to the hilt (no, I don't mean run slowly) and have some form of active homeostasis - a heat pump that can transfer heat in or out as needed, for example. Under most conditions, a very passive form of homeostasis is sufficient - have cold air or a cold fluid pass by hot components. That's fine, because heat won't generally flow against the gradient, so the temperature of the air/fluid is the minimum temperature the system can ever reach.
When you're varying the amount of heat you're generating, but the amount you're losing is fixed - particularly if the ambient external temperature is too cold for one or more components - then that is useless. The system will sometimes run too hot and sometimes too cold. That's not good. In those cases, what you want is somewhere you can dump extra heat you don't want when the system runs hot, and somewhere you can pipe that heat in from if the system starts getting too cold. Then you can always keep things just right.
The long and the short of it is this: It all depends on circumstances. Not all cold is created equal, nor are all machines the same.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
The water would not condensate on the CPU itself ...
make sure that water that condensates on the (cold) case
Aaugh, I can't stand it.
"Condensate" is a noun, it's what you get when water (or other liquid) condenses (verb) on something cold.
-- Alastair
The breath of the workers in the warehouse can only increase the relative humidity of the -18 C air to a maximum of 100%. Even if the -18C air is at its dewpoint, and therefore has the maximum possible amount of water vapor in it, when it is heated to 20 C insde the computer, the same amount of vapor will only represent about 10% of what is neccesary to reach the dew point according to this graph, completely preventing even the slightest chance of condensation.
Digital logic generally copes well with the cold, even if the manufacturer only promises down to zero Celcius (freezing point of water).
The real problem is water-based aluminum electrolytic capacitors. They rely on liquid water for their electrical properties. Go below freezing and the capacitance drops by ~80%. Essentially all commodity computer equipment uses these caps in the power converters. If you take them below freezing, the power supplies flake out. Long-term reliability will be crap, even if they "seem to work".
The advice in another comment is right: All the conventional servers and routers must go in a temperature-controlled room.