[Techtalk] Temperatures in server rooms

Anthony de Boer adb at adb.ca
Thu Apr 26 01:42:48 UTC 2012

Magni Onsoien wrote:
>  ... It is important
> to note that their temperatures are measured by reading SMART-data from
> the HDDs, i.e. are the temperatures on the HDDS, inside the cabinet, not
> the temperature of the air, which is what we'd usually measure when
> talking about the 20-25C temp. So it's also important to know the
> difference between the temperatures inside the cabinet and in the room
> as such.

One consideration is cooling consistency.  If everything was
consistently 25 or 30 degrees you'd be happy, but a lot of designs
can trap hot spots, and if you have to keep your room chilled to
15C to keep critical bits under 40 then there's a problem.  You want
airflow to go where needed to avoid hotspots, and not freeze spots
that don't need it.  Condensation issues are best avoided too.

Old-fashioned design with a false floor to act as cable duct and
air-conditioning plenum, with air going up through each rack, weren't
terribly efficient.

Current thinking seems to be "hot aisle / cold aisle" where the rows
of servers are front-to-front and back-to-back, with partitions over
each row and closing alternate row-ends, and unused rack spots covered
over too, so you have essentially one room with the fronts of all the
machines and the other that sees the backs, and you pump cold air into
the one and it has to go through each machine equally.

The days when air conditioning was a new thing and you had to
half-freeze people to impress them are long over.  Being able to keep
the room in a comfortable range for the people who have to work there
is important for morale and shortcut-avoidance.

Anthony de Boer

More information about the Techtalk mailing list