response 1 of 61:

Mar 4 05:45 UTC 1996

I think this is a good idea, but I am concerned about the long term
cost of running this. Our power expenses are high and going to get higher.
If this can be done without reducing the efficiency of our power consumption,
then it is clearly the direction we should be heading.

We aren't running on anything like a $12.95 surge suppressor, though. We
are running on a power conditioner. Others have the specs on this.

We also have some money pledged for a UPS already. I'm not sure how much 
is pledged and collected at this point.

response 3 of 61:

Mar 4 15:17 UTC 1996

   Thanks for entering this Mike.  One of the things we need to to is
figure out what capacity we want.  Right now I am making the assumption
that we want to stuff everything on it, plus some future stuff.

   "Future stuff", off the top of my head is A) a Sun news machine,
B) a mail machine, C) ?.

   This of course means that we need a BIG UPS.  What what constitutes
BIG, numerically I can't say yet.  This means that we need at least $500,
and possibly as much as $1000 for the task.  But once we've done that,
all of Grex can survive one of those disgusting summertime power outages.
   We need to come up with more complete numbers but I think we're
looking at about 3.3Kva or so of capacity--but thats just a guess.

response 5 of 61:

Mar 4 17:11 UTC 1996

Did we install a branch circuit with an ampacity of 30? [I love NEC
jargon....] 

response 7 of 61:

Mar 4 18:25 UTC 1996

Does the sun4 use less power than the sun3?  I agree that a possible future
news server is no reason to wait for a UPS big enough to handle it.  

What would be the disadvantages of leaving other components besides grex and
the modems off a UPS?

response 9 of 61:

Mar 5 06:50 UTC 1996

I was hoping for a point like (1) in #8, because otherwise the hidden cost
could be huge. I support this idea, then.

response 11 of 61:

Mar 6 04:19 UTC 1996

Would we create the equivilant of the third type of UPS you talk about,
Marcus, by plugging our power conditioner into a glitch mode UPS, or would
that create other problems?

response 13 of 61:

Mar 8 07:53 UTC 1996

Marcus, your information on UPS's is out of date.
1.) First off, the type of UPS you describe in #1 is variously described
    as an "on line", "continuous", or "true" UPS. Except for the most
    demanding applications, this type is not really needed. In this type
    the 120Vac line voltage is stepped down to a high current 12, 24, or
    48VDC level. This DC voltage is simultaineously applied to the 
    batterys and to the inputs of an inverter. The inverter converts
    the DC voltage back to a 120Vac sine-wave output. The advantage of this
    design is that it represents the ultimate in isolation, the output is
    separated from the input by 2 transformers, plus there is 0 cutover
    time. If the input 120Vac goes away, the batterys stop being charged
    but the inverter continues to see an uninterupted DC level at it's
    inputs. The disadvantages of these units are:
    1.) They are much more expensive because they need beefier electronics.
        Unlink other types, the inverter is in continuous operation. Unlike
        standby units, that only need a small AC to DC stepdown unit to 
        provide the battery with a few amps of charging current, these units
        require the AC-to-DC converter to provide sufficient DC amps for
        full-load operation. This requires more/heavier/expensive electronics.
    2.) The estimate of 10% power overhead for these units is wildly
        optimistic. The best of this type waste 30% in overhead and some
        are as high as 50%. 
    3.) The overhead created in #2 is disipated as heat. This units make
        a fair amount of waste heat. The prime factor that ages the
        batterys in a UPS is heat. The warmer they are, the faster they
        degrade and the sooner they need to be replaced. On-line UPS
        batterys typically have only a 2 year lifetime. Sealed UPS 
        batterys are not cheap to replace.
    Because of all of the above, on-line UPS are generally only required
    by really sensitive applications. Like maybe for labratory monitoring
    equipment or certain medical equipment.

2,) What Marcus describes in his #2 as a "glitch-mode" UPS, doesn't
    really exist as a real type. Possibly as a comment on really-bad/cheap
    implementations of "standby-by" units I'll describe below, possibly as 
    a description of an older type that no longer exists. Long ago, there
    were devices that provided backup power, but they had very slow cutover
    times. Typically they were contructed with a relay that had to close.
    This would result in a "blink" during cutover. Cutover time was anywhere
    from a 1/2 second to several seconds. These units were never designed
    with computers in mind, but more for electronic equipment that needed
    to be kept running even during a blackout. Nobody intentionally makes
    this kind of UPS anymore and they are certainly never sold for 

    computer use, as they would be worthless.

3.) The other basic type is what is known as a "standby" UPS. There are
    actually about 4 or 5 distinct strategys for achieving this effect,
    but generally this design works by running the output, under normal
    conditions directly from the 120Vac input. The 120Vac input is
    ussually filtered and passes through various circuitry, but the inverter
    is ussually switched off, and a low-current AC-to-DC converter is
    used to keep the battery charged. When the input AC drops below a
    predetermined level, the inverter is activated and the load is switched
    from the input line to the inverter. This switching time is what
    Marcus is reffering to in his "glitch-mode" above. However, except for
    really low-quality/cheap/bottom-end units, nearlyt all modern UPS's
    measure thier cut-over time in milliseconds, typically around 5ms.
    This is less that 1/2 of a 60Hz AC waveform. Almost all modern UPS's
    now sold are for computers and if someone made a unit that didn't
    switch fast enough, they'd quickly be out of business. I recently
    read several different reviews of UPS tests. None of the tested units
    switched slowly enough to cause a computer to glitch.

So the standby type is really what we want. Cost is affected by what you
want in the way of size and extras:
1.) Total KVA rating. The higher the KVA rating, the higher the cost. The
    KVA rating is similiar to Watts and specifies how much equipment can
    be plugged into the unit and operating simultaineously.
2.) Battery runtime. More batterys, the longer the unit will run at load,
    the longer of a blackout it can ride out, the higher the cost.
3.) Sine-wave versus square-wave output. The better/more-expensive UPS;s
    put out a 120Vac sine-wave that mimics as nearly as possible the 120Vac
    present at the wall outlet. Cheaper units will put out a square-wave.
    There are some units that put out a "modified square-wave" that is sort
    of a compromise between the 2. They break the square-wave into several
    discreet steps that sort of approximate a sine-wave. This is an area
    with a lot of debate. Some electronic equipment seems to run just
    fine with a square-wave input, others not. The main argument for
    sine-wave is that the engineers who designed the equipment were
    expecting sine-wave and it becomes a crap shoot as to whether a
    specific piece of equipment will do the-right-thing with square-
    wave input. I agree with this arguement.
4.) Cutover methodology. There are many different designs. Some more
    costly than others. Each with their own proponents of why their method
    is better than the compition. Best has their Ferrups system, APC uses
    a technology they call "line interactive", Unisom uses a system they
    claim is an "on-line/Standby hybrid", it really isn't.
    Caveat Emptor on this item.
5.) Better/pricier units have more/better filtering on the 120Vac line.
    Several of the top end units watch the line for low-voltage or
    high-voltage swings and cut into backup mode to keep the output at
    120Vac. A couple of other brands have a system that keeps the output
    steady at 120vac through an input range of 108 to 132Vac, beyond this
    range, the backup cuts in. The APC SmartUps series has this feature.
6.) Fancier computer monitoring. The more expensive units have more
    complicated computer monitoring capabilities and programibility. For
    Example, most modern units do NOT switch back to the AC line teh instant
    AC power is restored to guard against the possibility that the AC line
    may not yet be stable. This would result in needless rapid switching
    back and forth between backup and on-line states. These units have
    built in delays that wait a minute or 2 for the AC line to stabilize
    before switching back. The amount of time and the line stability
    factor is programable on some units.

response 15 of 61:

Mar 9 03:04 UTC 1996

Thanks, gregc. That is a keeper.

response 17 of 61:

Mar 11 02:37 UTC 1996

Trace is the name to remember for the ultimate in inverters.
 
How about connecting one of Tesla's poly-phase motors to and old 
exercise bike and using it to generate power for Grex?  ;)
 
Oh!  There we go!  9.5 KW 240 vac / 120 vac, 3600 RPM generator (No
motor) for $580 form Northern.  ;)

response 19 of 61:

Mar 11 12:31 UTC 1996

No.  It was already decided, quite a while back, to use hamsters (or possibly
guinea pigs) on treadmills for that.  The big selling point was the
possibility of training them to defend the circuit breaker box as well, if
I recall correctly.

response 21 of 61:

Mar 12 23:35 UTC 1996

Marcus - I think a clean shutdown should be a primary goal of this
project.  Given the length of some of the fsck error logs I've seen...
And in fact that's what inspired me to enter this item.

The amount of time that the system can last on battery power is 
a good question.  Should the shutdown process start within five minutes
after the cutover to battery?  Ten minutes?

We're in an interesting arrangement because it's actually possible
to have users who still have power while the system doesn't, which isn't
usually the case in most UPS installations I've seen.  The "power is
out, shutting down" messages have the possibility of being seen by the
users.

response 23 of 61:

Mar 13 03:30 UTC 1996

   If the system could last five minutes on the UPS, then we could
  - run the system for 1 minute without doing anything special;
  - after one minute warn that the system will be going down in
    another two if the power does not come back.
  - warn people about an eminent shutdown 1 minute beforehand.
  - actually shut the systems down if we don't get power back.

   Obviously, more than five minutes would be better.  I remember
reading something from Detroit Edison which said that more than
50% of their outages were 20 seconds or less, and about 10% were
longer than one hour.  Not sure about that last 40%, but I have
seen the power go out briefly several times now at the Dungeon,
only to start back in within seconds.  So a five minute UPS
provided grace period would stop probably half the power outages
from hitting us at all, and would let us bring the system down
gracefully for the longer periods of outage.