Re: UPS for node 0

["David Buckley" <db@xxxxxxxxxxx>]

I've had some experience in specifying a power system for a data
centres, my most recent for a 120KVA with a 24x7x365 requirement.
So far (a couple of years and counting) its performed flawlessly,
despite several utility interruptions and other provocations.  Its
not perfect (cost and sensibility constraints), but its pretty good,
and I believe it will continue to operate as required, despite it
being imperfect.

Its not at all obvious how to build a high reliability power system
for a data centre, as evidenced by many of the supposedly non-
stop "tele hotel" facilities in London and elsewhere that despite
investing heavily in power infrastructure, have non-the-less had
interruptions.  I did a design for a tele hotel (which floundered
due to the .com crash) with 24 * 365 * 20 years design availability,
and that requires considerably more infrastructure and investment
that any of the facilities I've looked at.

Many of the problems of keeping a data centre up also relate
directly to a node zero scale problem.

In specific answer to some point other folks have made.

1) UPSs in each rack - in commercial use, a really poor idea.  Small
UPSs are not adequately reliable, almost always are unmonitored so
you dont know they are about to fail, or even that they have failed
(other than a server has fallen off the map), and when they fail,
you loose the stuff it powers.  Unfortunately, a number of
manufacturers of rack based servers, of whom Compaq were probably
the best known thought it was a good idea.  If you only have one
rack in a non-controlled environment, it is probably appropriate (ie
it may be better than nothing), but if you have several, its bad.

2) Faults.  Semiconductor UPSs have very limited abaility to
generate enough fault current to flip a breaker, much less burn out
a fuse.  This is a problem with big UPSs.  So what chance does an
APC 1KVA UPS have - absolutely zero. This (of course) is
unacceptable, so what happens is when the UPS electronics detect an
overload they flip the static switch from UPS power to street power,
as street power does have enough grunt to burn out a fuse.  The
corollary of this is that its hard to clear faults that occur when
the prime power source is absent...  And that is why its helpful to
have a oversized backup generator - more grunt for the instant its
needed.

3) Earthing of racks - not strictly this topic - advice is to
earth.  It does ensure that your rack isnt an aerial radiating muck,
and it is a sensible safety precaution.  I've not got my regs handy,
so I dont know what the official position is.

The Big Question for the typical node zero owner is - "how much
trouble is it if I loose power?".  If you've wired your node zero
supply correctly, the most likely cause of node zero loss of power
is street supply failure.  If street goes, then the rest of your
home has gone as well, so there is little left for node zero to
operate, except that stuff which is supplied directly by node zero
power.

If you are in an area with frequent power outages, then you may
benefit from UPS cover to critical equipment.  If you power drops
once every five years, you probably wont.

if you have a UPS, then the UPS is your new weak point, allow for it
in your design.  Have two independent power feeds to node zero, 20A
each should do :-), and run one through a UPS.  Run critical loads
off both supplies.  For PCs that means PCs with twin power inlet
sockets, like many "proper" servers have.  For low voltage
supplies,
use pairs of (expensive) switched mode power blocks, run one from
each of your UPS and non-UPS feeds, and parallel their outputs using
diodes.  Fuse DC distribution, and breaker AC distribution.
However, the issue of fuse and breaker discrimination is difficult
at low current levels.  The 20A breakers feeding the radials should
be slower (type 2 or 3) than the breakers feeding individual
things.  If they are integrated RCD, then thats good too from a
safety perspective, but not from a reliability perspective.  This is
a risk tradeoff, make it wisely.

Your system design (assuming no generator backup) should be
something along the lines of "wait a few minutes to see if the power
comes back, and if it doesn't, institute an orderly shutdown".  The
two big questions after that are how to reliably switch power (so
you can turn things off) without compromising the reliability of the
power supply, and how to come back up.  The second question is hard
and system dependent, and may be as simple as "I'll do it
manually".  The former - the least bad answer is impulse (a.k.a.
latching, or pulse) relays, which keep their state without power.
These can be ransacked froman X10 appliance module.  You dont want
to trust your node zero power to X10, that is a reliability concern!

You also need to monitor that all your supplies are operating
correctly, otherwise you may only find something has gone when a
redundent feed fails...

Stuff like alarm panels which have inbuilt batteries it is best to
run them on their own circuit without UPS cover.

Need to be certain of a single source being up and got lots of
money?  A baby Cyberex may be just what you need:

http://www.cyberex.com/members/products/sts/compact_rackmount_euro/in
dexeuro.html

Connect between supply A (UPS and generator backed) and supply B
(street, in case supply A fails) and you will get highly available
power at the point of use.  $4000 IIRC.





UK Home Automation Meet 2004 - BOOK NOW!
http://www.ukha2004.com

http://www.automatedhome.co.uk

Member Offers - http://www.freeranger.co.uk/ukha