[Beowulf] Re: UPS & power supply instability
kewley at gps.caltech.edu
Fri Sep 30 13:53:16 EDT 2005
On Friday 30 September 2005 09:50, David Mathog wrote:
> RGB mentioned that "the HVAC is a nasty load". That got me
> thinking - when you increase the number of nodes running the
> HVAC servicing that rack(s) is going to sense the increased
> heat and compensate, presumably by running its motors more.
> So you could have an indirect problem such as:
> increased load ->
> increased HVAC activity ->
> UPS instability
> Where the load correlates with the instability
> but is not the direct cause.
> Does the HVAC cooling the test rack run off the same UPS
> as the rack itself???
Maybe I wasn't clear enough. We have two UPSes, one for computers only, and
one for HVAC only. Here is more detail of the feeds into the room:
substation A -> panel A --> UPS B -> HVAC control, fan, condensate pumps
\-> HVAC heaters, humidifiers
substation C -> UPS A --> PDU1 --(45 whips)--> 15 racks
\-> PDU2 --(48 whips)--> 16 racks
(leaving out some breaker panels)
We also have a feed, likely from substation A, into some wall outlets that
we don't use normally (used occasionally for vacuum cleaners, power tools,
rarely & temporarily for terminals, etc.).
> If so, it might be interesting to shut off that one HVAC, crank
> up the others in the room, maybe put a few fans in (on different
> circuits) to push the hot air over to them, and then try
> increasing the number of nodes again.
> That should at least rule out "lots of HVAC activity" as being
> the (sole) source of the problem.
The HVAC and the computers have only the following circuit elements in
common, to the best of my knowedge:
a) the main building feeder (upstream of substations A and C)
b) ground connections, in a loopy ;) topology
I think it's unlikely that there's significant feedback between the power to
the HVAC and the power to the computers, but feel free to point out
possibilities that I'm missing.
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
More information about the Beowulf