[Beowulf] Servers Too Hot? Intel Recommends a Luxurious Oil Bath
Ellis H. Wilson III
ellis at cse.psu.edu
Wed Sep 5 11:57:13 PDT 2012
On 09/05/2012 02:46 PM, Douglas Eadline wrote:
>> I'm not sure that google actually does servicing per se.. they mark it
>> dead, and just move on. The cost to service (or even to diagnose) is
>> probably higher than the cost of just overprovisioning.
> I would assume this as well. Over provision for an expected
> decay rate and life cycle. Colonizing insects figured this
> out already.
Not sure if they still do it, but in the video I shared they clearly
demonstrate a tech replacing a unit (kind of, he put a unit in where a
slot blocker was, but they didn't say whether that was previously empty
or was from a pulled unit). I have doubts that they try to fix all (or
any of) those bad, pulled units, but I bet to try and take maximum
advantage of the cooling and switching expenses per container they do
swap them out in chunks, maybe when some threshold is hit on a container
(10 or more machines dead or some such thing).
More information about the Beowulf