Slashdot Mirror
Oil Soaked Servers Coming Soon
grease_boy writes "A UK company will start selling server racks submerged in oil baths within a year. Very-PC is working on prototypes and says that because oil transfers heat more efficiently, power usage can be cut by fifty percent."
HWSpirit did a proof of concept here. I wonder if these guys were inspired by that.
But it's a decent idea. Oil has a high thermal capacity and will circulate through convection keeping the temperature down. Repairs and upgrades aren't going to be all that pleasant but some swarfega will get the grease of your hands after changing the motherboard.
However, the main problem I see is connectors. Existing connectors have been developed to work in air, except for a few exotic types. Watertight connectors are designed to work with wet environment outside and dry electronics inside, not vice versa, but in any case existing technology would require standard connectors to be used entirely submerged in dielectric. Modern connectors have much smaller contact surfaces than they did even ten years ago, and the distance liquid would have to move by capillary action before breaking the contact is quite small. It's hard to see how you could do accelerated life testing for such a system, which means it could be many years before we know whether they are reliable or not.
I recall when doing research involving electronics in Fluorinert we had to make soldered connections in liquid. Contacts that were frequently made and broken could be pressure contacts, but that is quite different from the situation in a server. And if we had known of a cheap substitute for Fluorinert we would have used it. The majority of oils degrade quite interestingly - you wouldn't expect bacteria to live in them but they can and do if the conditions are right.
These guys may have a workable solution to all the problems, but I can't help thinking that technology will make the concept obsolete. How does the performance of an old Fluorinert-cooled Cray stack up against a modern server in flops and GBit/s of IO per watt? (Hint: Don't bet on the Cray.)
Pining for the fjords
I know it was a joke, but I fear that, by replacing air by oil, the weight of a server rack might be a problem if it is not located in the basement.
Anyhow, even by reducing the power requirements by using efficient passive cooling to evacuate heat from the chips to the room, you still need to evacuate heat from the room.
imagine a beowulf cluster of these.
09 F9 11 02 9D 74 E3 5B D8 41 56 C5 63 56 88 C0
XXX#######
The only problem is that oil is a good solvent. Of course, computer equipment is obsolete in 3-5yrs, so maybe it's not an issue. However, the article mentions they tried motor oil first, so I wonder how much they actually thought this through. Motor oil, among other things, is much more viscous than traditional dielectric fluids. The fluids used in transformers are more like water in terms of viscosity. Lower viscosity provides better heat transfer. Also, since high dielectric strength is not an issue, I've got to think that there are some less-corrosive alternatives that will do the job without destroying the components. Half-baked at best, I think.
Exactly, the website and business proposal seem very amateurish. They actually tried using motor oil before realising that by some strange, arcane and entirely unpredictable process it correded PCBs? Am I being too presumptive in assuming that these people know very, very little about electronics?
Why on earth they didn't at least think to use highly-refined mineral oil like transformer oil is beyond me. I mean, filling a server with motor oil? Are you kidding me?
Someone saw the Tom's Hardware cooking-oil-cooled-PC experiment that was published a while back, and saw an opportunity to make some money. They didn't realise that Tom's Hardware used oil because it was headline-grabbing, cheap, easy to purchase and --oh yeah-- wasn't being used to cool a server that had to be stable and reliable. That doesn't mean it's the best choice of coolant.
Hell, you could do it with purified water if you wanted to, but your uptimes might take a hit.
http://www.theregister.co.uk/2002/05/11/bofh_and_t he_vax_cluster/
Didn't IBM use oil-cooling on one of their mini/mainframe computer systems back in the day? I seem to recall hearing stories of low oil indicators on the machines. Unfortunately my Google searches on the subject are coming up dry...
Shameless plug for my photos on Flickr
Fluorinert's ozone depleting.
Novec's a greenhouse gas problem.
Every other fluid in this class has the same set of issues, unfortunately.
They may be "clean" and non-toxic, but they're decidedly NOT environment friendly compared to oils-
and they're a hell of a lot more expensive than oils and not as effective at cooling things.
The reason why the fluids are used in the supercomputer industry is more the mess caused by the oils
on everything- and they're actively cooling the systems. Oils are actually superior to the fluids
in heat-transfer terms- it's why you have oil filled transformers for power distribution instead of
dielectric fluid filled ones. The specific heat of Novec is actually less than air's- the only advantages
these fluids have is that you can effectively move a LOT more of it quickly over a surfaces being cooled
without noise and you can refrigerate the stuff to below ambient to temperatures close to the freezing
point of water without condensation risks.
Oils tend to have issues with active cooling. Unless you're implementing vapor-phase, stirling cycle,
or aggressive peltier active cooling below ambient, you are actually better off with oils than the fluids
as they won't work as well at cooling- you'll be better off with air cooling.
This has been discovered by the overclock crowd and they have done a handful of oil-immersed PC's.
The main reason why you don't see a lot more of those oil immersed PC's is oil wicking
by the wires. Each point where a connector would be or a peripheral like a CD/DVD or hard disk is
hooked in has wires coming out of the system that will wick the oil or dielectric fluid out all over
the place. In order to deal with this specific problem, you'd have to resort to specialized sealed
header and other connectors for each edge case for SATA/PATA, VGA/DVI, etc. Those don't come cheap,
so the overclocker crowd tends to just resort to fishtank and similar plays for lan parties or
PAX/QuakeCon, etc.
So, in the end, it is a mixed bag. The oils are messier, but are actually more environmentally friendly
than the dielectric fluids- and they have a higher heat capacity and thermal conductivity in many cases.
I am not merely a "consumer" or a "taxpayer". I am a Citizen of the State of Texas
Slashdot ran a story on total-immersion computing using an oil bath, oh, four or five years ago now. He was using mineral oil. This is not to say it's a bad idea - on the contrary, it's rather overdue on the technology front. However, it does take about this amount of time to go from hobbyist to early market, so maybe this story should have been expected some time this year or next.
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)
I remember a guy called Dr. Ffreeze who submerged a motherboard in oil to push the limits of overclocking even further. So this isn't a new idea. His website describing his project is http://drffreeze.net/. The site doesn't look like it's been updated in a while, but it still contains info about what he did.
Half the power is consumed by cooling? Not likely. Given the current air conditioning technology, the coefficient of performance (Joules of heat moved by 1 Joule of air conditioning energy) is about 2.5 to 1. Assuming their system requires zero energy (It doesn't. They propose using a refrigeration unit to create convection currents) the best savings they could hope for is 28%. In reality, the heat needing to be moved will not go down. The energy savings will be produced by moving the same quantity of heat with a smaller volume of working fluid (oil instead of air). Some savings will also be realized by not having to condition the entire data center building to suit the equipment requirements.
Have gnu, will travel.