[Beowulf] Re: UPS & power supply instability

Robert G. Brown rgb at phy.duke.edu
Fri Sep 30 07:29:05 PDT 2005


David Mathog writes:

> For the experts out there: could a huge iron pipe, or a lot
> of iron rebar, produce this much of an inductive load if these were
> located just below the floor, say 6 inches from the power
> lines running to the racks?  The lab in question is on
> the bottom floor of the building so there could conceivably 
> be something like that underneath it.  I'm thinking not, but
> then I've never worked with the kinds of currents that are present
> here.

I don't think so -- never heard of such a thing, really.  Also, I really
doubt that any of the PF measurements are going to miss an actual phase
angle from real induction or capacitance on the supply lines, just as I
wouldn't be surprised if ALL they return IS the phase angle (which
presupposes a monochromatic system, at least in the derivations that
define the phase angle and power factor in the first place in any
standard physics or EE text).  Third harmonics are subtle, phase shifts
are not.

Also note the following -- suppose that there is enough neutral current
to cause roughly a 3V ripple at 180 Hz (and higher fourier components)
on the neutral line from IR between where the three phase load runs into
the common neutral and the point where the common neutral finally
reaches a reliable ground (e.g. building steel).  How high this voltage
really is depends on three things:

  a) The length of the run of the common neutral to ground (L);
  b) The gauge of the wire (A);
  c) The current (I);

(Ok purists, four things but I'm assuming copper as a given).  No need
to do the math, online tables abound:

Copper wire resistance table

AWG   Feet/Ohm  Ohms/100ft  Ampacity*   mm^2   Meters/Ohm  Ohms/100M

 10    490.2       .204        30      2.588    149.5       .669
 12    308.7       .324        20      2.053     94.1       1.06
 14    193.8       .516        15      1.628     59.1       1.69
 16    122.3       .818        10      1.291     37.3       2.68
 18     76.8       1.30         5      1.024     23.4       4.27
 20     48.1       2.08        3.3     0.812     14.7       6.82
 22     30.3       3.30        2.1     0.644     9.24       10.8
 24     19.1       5.24        1.3     0.511     5.82       17.2
 26     12.0       8.32        0.8     0.405     3.66       27.3
 28     7.55       13.2        0.5     0.321     2.30       43.4

These Ohms / Distance figures are for a round trip circuit.
Specifications are for copper wire at 77 degrees Fahrenheit or 25
degrees Celsius.

SO we node that 12 gauge wire (a likely enough candidate for a shared
neutral) is roughly 0.16 Ohm/100 ft (note round trip factor of 2 in this
particular table).  For lack of data I'll assume that each phase is
breakered at 20 Amps (note "Ampacity"), which actually assumes runs of
<100 feet IIRC (otherwise you have to bump to the next larger wire in
code IIRC).  20x0.16 (peak) = 3.2 volts per 100 feet.

Of course with LUCK one's common neutral runs are both shorter and
thicker, although electricians dislike running 10 AWG because it is a
pain to bend and pull through conduit.  More often they'll double up a
20 on a shared neutral with an inductive load on a longish run.  Another
common server requirement is to keep the neutral runs SHORT -- I recall
reading < 25 feet somewhere on the net last time I researched this --
which would drop the neutral line IR (peak) to < 1V.  My KAW has
measured every bit of 3V in our server room, though, on a line being
driven at maybe 80% of capacity with its own neutral (noting that this
includes BOTH parts of the voltage drop -- supply line too).

The three phase common neutral current from non-PFC supplies doesn't
increase the peak current, so the neutral line ripple remains in the
(say) 0.5-3V peak range depending on how far your common neutral has to
go to find building steel or at least a conductor with "zero resistance"
tied eventually to building steel (which we will assume is dead solid
ground compared to the 120 V peak on each phase coming out of the final
transformer, although without knowing the details of how the final
transformers are wired this might not be correct).  It carries 3x the
neutral current of a single line (instead of NO current as would be
expected by a purely resistive balanced load) but that's only because
its FREQUENCY is 3x greater -- the peak remains a single phase line's
peak, breakered at or below 20A.

I think that this makes me mistaken (yesterday) about the line heating
-- it is at most 3x the line heating of a single neutral carrying the
same load, not 9x, because the resistive heating produced by each phase
is effectively independent.  Still quite enough to make the neutral line
hot to the touch where it should be stone cold, because I^2 R = VI = 20A
x 3V = 60W per phase per 100 feet (really x 0.707 -- this is order of
magnitude stuff).  Times three for the third harmonics on a common
neutral.  So your neutral line contributes anywhere from perhaps 50
watts to over 100 watts (per phase) to room heating inside a fairly
small volume of copper, and is basically thermally insulated by the wire
insulation so that it gets hotter still.

I go into all this detail to help with the Liebert discussion, as in:
"how long is the common neutral run from where the three phases are tied
together on the load side to where they finally reach a solid ground?"
"what is the measured voltage drop across this leg?"  "if you put a
scope across this line to ground (e.g. the ground wire or possibly the
rack chassis, which should be tied to the ground wire I'd guess), do we
see 180 Hz ripple with the scale set and triggered in the 0-5V range?"
"how hot IS that common neutral to the touch when we crank up the load
as far as we can (but under the circuit capacities)?"

To sound really nifty cool and ask all of these questions at once, you
can phrase it as: "Could this problem be caused by ground loops
interacting with the third harmonics generated by switching power supply
loads?" as that's the proper name for what this whole discussion is
about (although an unusual kind of 180 Hz ground loop -- they are more
commonly the 60 Hz ripple you hear amplified in e.g. amplifiers and
common electronics caused by differential voltages as objects are
connected by wires as if those wires had zero resistance when they
really don't).

To start learning about ground loops and the many, many problems they
can cause you can visit here:

  http://www.epanorama.net/documents/groundloop/

or (GIYF) any of a zillion other places in webland.  To quote one
paragraph from this:

  Common Causes for Computer System Problems

  Many times when a user thinks that his system is 'bad' or has 'gone
  bad' the fault is electrical or magnetic in nature. Monitor problems
  are very often caused by nearby magnetic fields, neutral wire
  harmonics, or conducted/transmitted electrical noise. Intermittent
  lockups of computers are very often the caused by a Ground Loop, an
  electrical phenomena that sometimes manifests itself when a system and
  it's peripherals are improperly plugged into different electrical
  circuits.  Many don't even know if their wall outlet is properly wired
  and grounded, an absolute necessity for a computer and peripheral to
  operate reliably and safely.

  Have you ruled out Ground Loops in your computer system ? Ground loops
  can cause problems to LAN connections if not properly wired. A ground
  loop caused by RS-232 connection to other computer can cause computer
  lockups.

Note that "improperly plugged into different electrical circuits" --
that is, different phases.  It is a bad idea and even potentially
dangerous to be careless about running multiple phases on tightly
interconnected hardware.  Where "careless" includes many things --
treating the ground loop problem lightly, but ALSO the risk of certain
wiring mistakes (like accidentally cross-connecting the actual supply
lines of two different phases through some arrangement of pieces of
equipment.

Note finally that UPSs and DC power supplies in general have unique
ground loop opportunities.  This is because (depending on how they are
engineered) they may provide a voltage between point A and point B that
is dead on some spec -- +5V, +120VAC -- but BOTH point A AND point B may
NOT be at zero potential WRT the two wires (ground and neutral) that
represent "ground" to your system from the point of view of both signal
isolation and safety.  These kinds of ground loops are very dangerous as
shorting the isolated "ground" to the real ground or neutral can causes
considerable current to flow.  As in copper vaporizing, fire causing,
spark blasting currents.  

Ground loops can even kill -- the reason code requires ground fault
protected outlets on bathroom, outdoor, and kitchen wiring is because
the NEUTRAL line of a high resistive load can carry enough voltage to
put a fatal current through you (a fatal current being only milliamps at
the wrong frequencies, where 60Hz is alas a wrong frequency!) if you're
standing on a wet bathroom floor in your bare feet.  Old pre-code house.
Wiring: two 14 separate asbestos insulated gauge wires running back 200
feet to a 20 amp fuse connected to these nifty ceramic insulators.  A
space heater plugged into the receptacle.  Unseen, the neutral wire
inside the receptacle is in contact with the metal of the shiny brass
receptacle plate and is floating at a few volts but with a LOT of
capacity to deliver current.  This doesn't bother your heater -- maybe
it is running at 110V instead of 120V, but that's still enough to make
it plenty hot.  You reach down to unplug it after your shower and
WITHOUT TOUCHING ANYTHING YOU SHOULDN'T BE ABLE TO TOUCH 30 mA is
diverted through your wet torso to ground, and the 60 Hz induces
defibrillation.  Your cooling corpse is discovered the next day.

Used to happen, used to happen.

So take ground loops seriously.  WHATEVER your problem ends up being,
dollars to donuts says it is a ground loop of SOME sort -- the major
question is where and what kind.

   rgb
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.scyld.com/pipermail/beowulf/attachments/20050930/9f459262/attachment.bin


More information about the Beowulf mailing list