[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[datacenter] peak capacity planning for power vs. cooling

Is this a reasonable statement to make:

When planning power capacity for a machine room, you need to base it
on the peak power consumption of the machines, usually when they are
powering up / booting, because it is possible that they will all power
up simulataneously.  However, when planning cooling, you can plan
based on average load, with some headroom, because even if everything
will be at peak heat output simultaneously, it will be a brief event,
and cooling averages out naturally over time.


For example, say I've got a lab that will host 100 units.  Based on
measurements, I know that each unit peaks at just under 0.8 amps
briefly when powering on.  When running under load with disks active,
it draws about 0.45 amps.  When up and idle, 0.3 amps.

For power, I must not exceed 20 units per 20 amp circuit (0.8 amps +
20%), and should only put 16 units per circuit.  For 100 units, I
should have 120 amps or more of capacity feeding the lab.

For cooling,
 peak: 100 x 0.8a x 120v x 3.41 BTU/h per watt = ~2 3/4 tons
 normal high load: 100 x 0.45a x 120v x 3.41 BTU/h per watt = ~1 1/2 tons

Two tons of cooling for this lab should be theoretically sufficient.
It is possible for heat output to exceed this amount substantially,
but only for a very brief period.  That's not okay when it comes to
power, but it is not a problem when talking about cooling.

Does this reasoning make sense?

  --  Cos (Ofer Inbar)  --     [email protected]  781-273-2380
  --  Permabit, Inc.    --  [email protected]  617-252-9600

To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]