UPS recommendations

[Trent Piepho]

On Wed, 13 Nov 2002, Dan Christensen wrote:
> I'm also curious about monitoring the UPS's.  If there are several
> of them, does one usually monitor all of them using the head node,
> or split this task up?

We have four APC smartups for our current cluster.  Each UPS is connected via
a serial cable to a machine that it is powering.  We don't have all four
connected to one computer for a few reasons.  The obvious one is that our
motherboards have two serial ports, not four.  Slightly less obvious is that
if one circuit goes down (eg.  trips breaker) but another does not, you could
have your UPS monitoring machine power down, but other machines on another
circuit still running, with a now unmonitored UPS.  And then when that UPS
also loses power, there is nothing to tell you about it or do a shutdown.

We use some opensource software called apcupsd.  I think now APC has their own
powerchute software for linux, but I'd rather have something opensource, and
apcupsd works very well.  We run a "master" daemon on each node with a serial
connection to the UPS, and a "slave" daemon all the other nodes powered by
that UPS.  When power fails, all the nodes will get shutdown if necessary. 
The daemon will email you for power outages and bad UPS batteries.  There is a
cgi script you use that will provide status for any number of apcupsd daemons.