[Beowulf] Re: Emergency Power Off

David Mathog mathog at caltech.edu
Mon Mar 19 14:06:10 EDT 2007

Jim Lux <James.P.Lux at jpl.nasa.gov> wrote
> At 06:52 AM 3/18/2007, Robert G. Brown wrote:
> >To answer my own question (GIYF, after all:-) there is a white paper
> >here:
> >
> >HOWEVER, if you love your local firemen and want them to live (or love
> >yourselves and the other employees who sit near the data not-a-centers),
> >the same white paper says that an EPO switch is still a very good idea
> >for small server rooms and wiring closets that are not "data centers"
> >but are just "data closets" or "data rooms that aren't quite centers".

That describes mine to a tee, some racks, some computers on shelves.

> Another way that folks avoid the code is by putting the UPS into a 
> rack with the equipment being powered.  That puts it on the other 
> side  the "line of demarcation" between that which is subject to code 
> and that which is "internal to the equipment" and subject to 
> different rules.  And, here, the regulatory environment gets very 
> tricky, depending on who's doing the racking and stacking.  In Los 
> Angeles county, for instance, an "assembly" like this would require 
> UL (or other recognized testing lab:RTL) certification or, possibly, 
> wet-stamped drawings by a PE in order to be legally 
> sold/transferred/etc.  So that means you can't just order it up from 
> your distributor already configured, unless they get the right certs. 
> {There are exemptions for R&D equipment, and things destined for the 
> movie industry, or things destined for a computer room (oops, now 
> you're in Art. 645 territory again)}

After reading all of this I don't know if we're within code or not with
respect to EPO, although I'm sure all the UPS units are UL certified. 
The engineers and electricians who rewired the room last
knew that it was going to hold UPS units, and knew their sizes,
but nobody (including inspectors) told us to install an EPO
for the UPS units.   These are all fairly small (1500, 2 x 2200,
800, 1000 - any one of which is more than I'd want to share a room with
if the sprinklers were running).  3 are on the racks, 2 are on the
shelves.  Unfortunately only two of them have separate EPO ports.
The room currently has an over temperature shutoff, and a big red
"KILL" button, both of which are wired the same way to throw the
main breaker in each of the two panels in the room.  Once the panels
are off it leaves the 5 UPS units in the room running.  That was
intentional - if the overheated most of the equipment would shut off
instantly and the servers would have a few minutes to power
off normally while running off batteries.  (It all worked correctly
a few months back when the main campus chilled water generator broke.) 
 The computers attached to the UPS units are set to shut down if
the power doesn't come back on within a minute, and that shuts off
the UPS units (inverter kill).  Assuming they are still running
normally, which they might well not be if the sprinklers have gone off.

After looking at the specs I believe that at least 4 of the 5 
can be wired to shut down instantly.  (Not sure about the one
desktop size APC unit in the room, as it isn't mine so I don't
have the manual). However it seems that turning off the
tripp-lite UPS requires a positive voltage (12V) on the
inverter shutdown line.  Which means, that even after the
power has been killed (overtemp, red kill button,
or building power shut down) the fireman entering this room would
still need to smack another (not yet extant) red kill button, and
that button is going to need DC power in order to work.  Googling
turns up other EPO installations requiring much higher DC voltages
than the tripp-lite needs, for instance here:


where they specify 56 volts for the EPO, but only 5mA of current when
the switch closes.

Clearly a second red kill button (EPO for the UPS units) would be a good
thing.  Presumably somebody sells an EPO control unit that
has all the pieces:

AC in
AC to DC conversion
battery or capacitor backup of the DC voltage
a bunch of ports to plug in lines to control the EPO devices

Who sells it?
How much?

Or is every EPO controller custom made???  


David Mathog
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech
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 mailing list