[Beowulf] Servers Too Hot? Intel Recommends a Luxurious Oil Bath

[Lux, Jim (337C)]

On 9/3/12 5:04 PM, "Christopher Samuel" <samuel at unimelb.edu.au> wrote:

>-----BEGIN PGP SIGNED MESSAGE-----
>Hash: SHA1
>
>On 03/09/12 23:15, Douglas Eadline wrote:
>
>> There was a previous discussion here. Curious to see Intel's
>> interest in this tech. I assumed it is mostly used for edge cases.
>
>It's going to cost a heap to cool an Exascale system, perhaps then the
>overhead of making bespoke racks for these things will make some sense.
>
>But even then it'll take a lot of clever engineering to avoid the
>issues that Jim mentioned; the capillary action of oil through
>insulated wires hadn't occurred to me - that could be very messy


Yes.. That's one of those learn by experience things... You silicone seal
the box...This time I won't get oil leaks.. You come back the next
morning.. What's this puddle of oil on the bench?

Then, there's the atmospheric pressure pumping phenomenon.  As the
pressure changes (actually temperature changes are a bigger pressure
changer (Gay-Lussac) than the weather, but both do it), either oil pumps
in and out, or air comes in and out, bearing moisture.   For low voltage
PC stuff moisture's probably not a big deal, but for HV insulation, it
sure is.

Basically, it's a lot like sealing vacuum systems. Everything leaks.


Then you thank the oil gods that it wasn't silicone oil, which is
impervious to all solvents and soaps.  Mineral oil cleans up fairly easily
with dish detergent. ( you don't have to use Lux brand.. I don't)


(this is why pneumatic systems are better than hydraulic.. Even if the gas
is springy, when it leaks, it doesn't make a mess!)