Temp at Level 3 data centers

Thomas Bellman bellman at nsc.liu.se
Wed Oct 11 18:56:31 CST 2017

On 2017-10-11 19:09, Naslund, Steve wrote:

> I would also be concerned because if they lose one of the a/c units
> what would the ambient temperature rise to?

It doesn't matter much if the "normal" temperature in your DC is 10
or 30 degrees Celcius; if the cooling system is barely keeping up
with that, and you loose half your cooling capacity, then temperature
will rise pretty quickly, until the servers are literally cooked (i.e.
temperature reaching 100°C or more).

The spare capacity of the cooling system is the important information,
not the starting temperature.  That difference of 10-20°C in starting
temperature will just give you a few minutes extra, not *save* you, if
there is not enough spare capacity in the cooling system.

Assuming a reasonably densly packed data centre, at least; with low
power density, thin unisolated walls, and winter outside, you might
survive even a full cooling failure. :-)

Also, depending on your cooling solution, a partial failure might not
be very common.  We have district cooling providing us with cold water,
and if that stops pumping water, and we use up our 30m³ buffer tank
(which is enough for 30-40 minutes), *all* cooling stops.  But on the
other hand, we have capacity enough to survive even if they give us
16°C water instead of the normal 9°C water.

> I would want them to tell me what the set point of the a/c actually is.

That I agree with.

> Bottom line 80 F input air is too hot in my opinion and apparently
> the equipment's opinion as well.

Unfortunately the default settings of most servers are not very well
thought through.  They will typically spin up fans *much* more than
is actually needed to protect the hardware, and often there is no way
of changing that for the user.  And if you try to get the manufacturer
to tell you what the most power-efficient inlet temperature is, they
will just tell you "oh, we support anything between 5°C and 40°C" (or
whatever their actual limits are), and absolutely refuse to answer your
actual question.

Thomas Bellman <bellman at nsc dot liu dot se>
National Supercomputer Centre, Linköping University, Sweden

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: OpenPGP digital signature
URL: <http://mailman.nanog.org/pipermail/nanog/attachments/20171011/c5bb8cd4/attachment.sig>

More information about the NANOG mailing list