Summary: UPS's (Uninterruptable Power Supplies)

[Peter S. Shenkin]

Now that I've gotten answers to my queries, they are more coherent if the
questions are phrased a bit differently;  therefore this summary is
organized around questions that are a little bit different than the ones I
orginally asked.

This summary is longer than I'd like it to be, but the excerpts I've
included do add information not included in my four- or five-line summary
for each topic, plus they give my original sources of information.

In my excerpts I don't quote everyone who replied; but I hereby thank all
who replied.

OVERALL SUMMARY:

A ballasted standby unit -- inverter normally off, but kicks in on
failure, with delay buffered by a ferroresonant transformer -- appears
adequate for all computational equipment.  A pure standby unit (no
transformer) will work for some computers.  True on-line -- inverter
always on -- appears unnecessary.

Q: Of the three commonly available types of UPS -- standby (with ca 4ms
   delay to battery startup following power failure), ballasted standby (with
   ferroresonant transformer buffering this delay time), and true on-line
   (inverter always on) -- which is the minimum level of protection necessary
   for computational equipment?

A: Standby works for most computers, but not always for other things like
   modems;  ballasted standby should always be adequate.  It was also noted
   that various machines (SparcStations and SGI PI's) seem immune to hardware
   damage due to the power going out suddenly (as in yanking the plug).  If
   this is the case, then a UPS is a frill, rather than an important
   insurance item.  Excerpts: [[ My interpolations are in brackets like this.
   -P. ]]

X-From: seibel at zeno.mmwb.ucsf.edu Organization: Computer Graphics Lab, UCSF

We have quite a variety of hardware here, including about a dozen irises.
We've had perhaps four or five power outages in the last few years.   The
interesting thing is that the irises have always come through with no
damage or data loss.   It looks to me like SGI builds a pretty solid
package, so if you're pressed for funds, running on line power may not be
that risky.

X-From: poage at sunny.ucavis.edu (Tom Poage)

We have a Sun 3/150 as our departmental computer.  I frequently see the
lights flicker due to power transients; PCs reboot or hang, dumb terminals
go whacky, yet the Sun never seems to be affected.  Overall I'm very happy
about this.  In contrast, our old departmental computer (a Z80 based S-100
bus system) was very sensitive and had to have a UPS.

X-From: "David W. Abraham" <DWABRA at IBM.COM>

I have a PI 4D/25 for which I bought a UPS from a company named BTG in
Vienna, VA.  (703)-556-6518.  It is their model 1200VX quality 1 , which
is I believe 1200 VA, for $906.00, plus $210 for the RS-232 shutdown
software.  It works fine, and is of the on-line sort as you have
described.  More grist for the mill...  [[ At this price can it really be
true on-line?  -P. ]]

X-From: guy at phy.duke.edu (Guy Metcalfe)

Sorry, no personal experiance to give, but in the oct and nov 1988 issues
of Byte magizine there are 2 articles by Mark Waller on this subject.  I
found them informative enough at the time to keep them around in a box as
reference.

X-From: karn at thumper.bellcore.com (Phil R. Karn)

[[ He comments that standby + Ferroresonant has disadvantage that it is
   large, massive and acoustically noisy, and that just standby is enough for
   most things in his actual experience;  true on-line clearly provides the
   most consistent power signal.  ]]

So, in general, SPS's [[ (standby inverter plus relay) ]] are just fine for
most computer equipment. And if you have more sensitive loads (modems,
etc), add a small ferroresonant transformer to protect them.

[[ I asked him by email if the above didn't mean that he simply built
something like the BEST unit himself;  and if so, why not start with a
BEST: ]]

No, it's not the same as buying a BEST UPS because you only need put
the ferroresonant transformer on the more sensitive loads; the switching
power supplies can be powered directly from the inverter. This minimizes
the size of the ferro that you have to buy, thus minimizing noise and
wasted power.  Ferros are not like ordinary isolation transformers that
you already use; they operate in the saturated mode, so they are quite
accoustically noisy. They are also less efficient.

X-From: eho at clarity.princeton.edu (Eric Ho)

[[ Eric Ho recently also asked questions about UPSs, and summarized
   to comp.sys.sun;  look there for the excerpts he quoted.  -P. ]]

Q; Why do you want the UPS to turn itself off after it brings the computer
   down?

A: After a power failure and bringing the computer down, the UPS batteries
   could drain if the power didn't come back up soon.  This would damage the
   UPS (or at least the batteries), and in addition would leave the computer
   with no protection should the power fail again shortly after it came on
  again.  Also, startup loads can be too high for the UPS to handle.

Note:  I did ask Best whether their unit does this, and the answer is yes,
it is programmed to switch itself off two minutes after it signals the
computer to go down.

Excerpts: [[ My interpolations are in brackets like this. -P. ]]

X-From: mk59200 at tut.fi (Kolkka Markku Olavi)

When your machine is "down" it's still ON and consuming power.
Discharging the accumulators in the UPS all the way may damage them, or at
least decrease their lifetime.  Switching the power output off saves the
accumulators.

X-From: don at zl2tnm.gp.govt.nz (Don Stokes)

Probably not much of an issue on a small system, but on a big one, once
you have pulled power out from under a system, you want to bring it back
up again in a controlled fashion.  The reason for this is that a startup
loads can be *much* higher than normal loads; some devices, eg large disk
drives, can pull something like ten times the normal running load.  Just
loading power supply capacitors can pull surprisingly large currents for a
short time.  An uncontrolled startup load on the VAXcluster at GPO stands
a pretty good chance of tripping the power conditioner out (we don't have
a UPS; 100kVA UPSs cost *real* money).

X-From: moe!paul%ppgbms at uunet.UU.NET (Paul Evan Matz)

In our application, we have to support power outages for long periods, and
many such outages in a single day (rural hospitals).  Any unattended
machine's UPS should be able to shut itself off after the machine has
halted.  Otherwise the battteries could conceivably be totally discharged;
In this condition, if the power comes up and the machine reboots, the
machine is totally vulnerable to power outages.

[[ Following is reply from email from me to him:  -P. ]]

>I just called Elgar, and I think your prices are a bit out of date....

The IPS 400+600 is available for $525 from one of Elgar's distributors.
(Call Bob Roth, Manchester Equipment, 516-434-8700).  Unisafe for SunOS
WAS $99, and is worth a bit less in my opinion.

> Also, these  and their
>smaller units are standby units, not online units.  This means that there is 
>a gap -- they quoted 4ms -- between the time the power goes off and the UPS 
>kicks on...

This gap (2-4msec, typical) doesn't bother a SPARCstation or 386i.  A 3/60
seems to be sensative to a sneeze.  I don't know about 3/50s.

>It also means that you get little or no power filtering or surge
>protection from the UPS.

I don't think this is true.  Their spec sheet shows noise, surge/transient
protection for the UPS's output.  I'm not sure what you mean by "little",
but assume it's not bad compared to nothing.

Q: What about this UL-544 spec (low leakage current) that Clary says they
   meet?

A: This has nothing to do with the UPS as a power supply;  it has to do
   with the UPS as a device.  It is a measure of how much current will leak
   to the chassis from the hot wire under controlled conditions if the ground
   wire is disconnected.  This spec is used for equipment used for medical
   and dental purposes.

Excerpts: [[ My interpolations are in brackets like this. -P. ]]

X-From: poage at sunny.ucdavis.edu (Tom Poage)

The Standard for Medical and Dental Equipment, UL-544, recommends a number
of different limits for leakage current, depending on the application.  In
general, medical and dental equipment are categorized into patient care
and non-patient equipment.

Leakage current, as far as your application is concerned, is defined as
that AC or DC current that "leaks" (via electromagnetic coupling through
insulation and air) from active ("hot") circuitry inside of the device
through the chassis and to ground through the third grounding prong (since
the ground prong is attached to the chassis).  There are also other
definitions that don't apply to you.

The limit for non-patient equipment is 0.5 milliamperes for frequencies up
to one kilohertz; this happens to be about the perception threshold in
humans for 60 Hz sinusoidal current.  The UL chassis-ground current
leakage limit for patient care equipment is 0.1 milliamperes.

I wouldn't worry about it.

[[ Following in response to email request from me for clarification.  -P. ]]

To perform this kind of test, the tester is plugged into an electrical
outlet, the device power cord is plugged into the tester and a measurement
wire is connected from the tester to the device chassis.  The plug ground
is then temporarily lifted (disconnected) and the leakage measured as that
current which flows from the device active circuit (the hot wire) to the
device chassis and into the tester wire and its measurement circuitry.
Some measurement levels are down to a few microamperes.  As you might
imagine this requires some sensitive circuits.

Peter S. Shenkin, Department of Chemistry, Barnard College, New York, NY 10027
(212)854-1418  shenkin at cunixc.cc.columbia.edu(Internet) shenkin at cunixc(Bitnet)