Power cut if temps are too high

Thomas Bellman bellman at nsc.liu.se
Tue May 28 13:12:59 UTC 2019


On 2019-05-27 18:18 +0000, Mel Beckman wrote:

> Before the trigger temperature is reached, the NMS would have sent
> various escalating alarms to on call staffers, who hopefully would
> intervene before this point.

Would they actually have time to react and do something?  In our
datacenters, we reach our cut-off temperature in about 20 minutes
if cooling stops.


> This system has triggered one time, successfully shutting down the data
> center on a holiday weekend when people missed their notifications, and
> undoubtedly saved a lot of hard drives. When we got to the room the
> temperature was over 115°, but the power was cut at 95°.

Presumably that was °F, not °C.

I have heard from people who did *not* have automatic cutting of the
power at high temperatures.  Their computer room reached 100°C in
places; some keyboards apparently looked like a certain Salvador Dali
painting afterwards...  (But I think they had very few actual servers
or disk drives breaking.)  The reason it didn't get even hotter, was
that as temperature rose, servers started overheating and shut them-
selves down, thus lowering power disippation more and more.


Our system for cutting power at high temperatures is part of the PLC
monitoring power and temperature in the computer rooms.  It sends a
signal to the large breakers connecting the power subcentrals (where
all the 16A fuses are) to the power rail feeding the room.  I believe
our PLCs are from Schneider Electric, but anyone who delivers PLCs
for controlling power and cooling in a datacenter should be capable
or programming their PLCs to do the same.  You just need to remember
putting it in the specifications when you contract the building. :-)


	/Bellman

-------------- 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/20190528/e36202a9/attachment.sig>


More information about the NANOG mailing list