response 25 of 61:

Mar 13 10:09 UTC 1996

Most modern UPS's that are any more sophisticated than the really bottom
of the line "bare-bones" units, ussually come with at least these 2 outputs:
1.) One signal that indicates that AC line power has been lost,
2.) And another signal that indicates that only XX minutes of battery
    power are remaining. XX varies from manufacturer to manufacturer, but
    is also ussually programable. This is ussually a 2 minute warning.

Older units ussually only had 1 output(if even that) that just signaled when
AC line power is lost. The problem with that is that you don't really want
to start a shutdown every time the AC line glitches. As STeve said above,
most power line failures are transients that only last about 15 seconds.
Even the smallest cheapest units have enough batterys for 5 minutes of 
runtime, so the idea is just to "ride through" the short power
interruptions.

The way I would handle this is:
1.) When the AC-line-power-is-lost signal was detected, send a message to
    all users saying something to the effect: "POWER FAILURE. System may
    start auto-shutdown in XX minutes, Please be prepared." The file
    /etc/nologin would be created at this time to prevent any more users
    from logging in.
2.) The system would otherwise continue to run normally. If AC power came
    back on, another message would be sent to all users: "SYSTEM POWER
    RESTORED. All is well.", and /etc/nologin would be removed.
3.) If power stayed off and 5 minutes[or a number that we determine through
    trail and error] had gone by, we send another message to everyone:
    "ATTENTION: POWER HAS NOT BEEN RESTORED. System will very likely shutdown
    in the next XX minutes. Save anything important you are doing NOW. This
    will be your last warning."
4.) When the second 2-minute-battery-warning signal from the UPS is
    dectected, we begin an immediate shutdown to single user mode. The
    system then waits for the UPS batterys to run out and for the UPS to
    shutoff power to the system.
4a.) (Optional) We could *maybe* give the users a final 30 second warning
    at this point, but that might be pushing the window of safe battery
    life in which to do a shutdown.
5.) We will have to modify single user mode slightly to run a program that
    continues to monitor the signals from the UPS. It is possible that AC
    line power would come back on after the system went into single user
    mode. That would prevent the OFF then ON pwer cycle that would cause
    the system to reboot normally. The program would monitor the UPS and if
    it detected that AC line power had been restored, it would put the system
    back into multi-user mode.
5a.) (optional) Alot of UPS's also have an input that allows you to signal
    them to shutoff. We could then install a small prgram at the single user
    level that does one last check to make sure that AC line power is still
    off and then it sends the Ok-I'm-all-shutdown-and-ready-go-ahead-and-
    shut-me-off-now signal to the UPS.

All of the above events would also be logged to a file.

response 27 of 61:

Mar 13 16:59 UTC 1996

A cold-boot from power on is automatic. Operator intervention is only 
required in the (normally very rare)case of when the file system is badly
screwed up. The purpose of the orderly shutdown is to *prevent* the
file system from becoming badly screwed up.

response 29 of 61:

Mar 19 07:09 UTC 1996

I recommend, and hereby nominate, that response #25 be implemented
by unanimous vote of the Board. It is well thoght out, totally coherent,
and demonstrates why experience can be a dear teacher, whether as an
historical event, or as a well considered thought-experiment, ala Einstein.
 
I would dump the 30 secnds option ... enough is enough. Ppl can count.

response 31 of 61:

Mar 19 10:06 UTC 1996

I agree.

response 33 of 61:

Mar 19 11:57 UTC 1996

Well, we've got a "plan the budget" meeting coming up this friday, so that's
a good time to discuss it.  

response 35 of 61:

Mar 27 04:46 UTC 1996

If anyone cares, I can pester my boss at accessUS about our UPS.  I've been
told that it's kept the network (including dozens of modems, routers,
portmasters, several copmuters, *etc*) up and running for 45 minutes, when
the power was restored.  I've personally seen the system ride through two
power outages without a glitch.  (The power source at our office can be flakey
at times . . . not unlike Grex's.  8^)

response 37 of 61:

Mar 27 08:10 UTC 1996

is anyone looking into costs vs runtime for UPSs? Myself, and just on the
runtime issue, i'd go for about a 15 minute window, depending on the
cost of course.

response 39 of 61:

Mar 27 14:40 UTC 1996

   It would be *great* to have a UPS that could keep us alive for
45 minutes.  Unforunately, I don't think we could afford it.  Well,
we could, but we'd have to sell everything we own to get it. ;-)

  Agreed that 15 minutes would be more realistic, and would prevent
us from harm in at least 70% of the instances we go through power
instability.

response 41 of 61:

Mar 28 02:32 UTC 1996

I'm not sure that we need even 15 minutes.  In other places I've lived that
would have been useful, but I really don't think so for Ann Arbor.  Power
failures in Ann Arbor, at least in my experience, last either a maximum of
two or three seconds, or several hours or longer.  I think that in general,
if we lose power for more than 10 seconds, even a 45 minute UPS probably won't
help us much.

response 43 of 61:

Mar 28 22:08 UTC 1996

   The amount of time that the UPS can feed everything is a function
of how many amp-hours of battery there are feeding the UPS.  There
will be some amount of time that the UPS can deliver its power at
maximum capacity; that is likely to be some fraction of an hour,
like a quarter, or eighth.

response 45 of 61:

Mar 30 23:54 UTC 1996

The APC "BackUPS" series are square-wave UPS's. We would want either the
APC "SmartUPS" or "SmartUPS v/s" series. The 280 is also a pretty small
unit. It's only reated for 280VA peak, or about 180watts.

response 47 of 61:

Mar 31 08:23 UTC 1996

My thoughts are that a cost/value analysis (not formal) could be done
on UPSs with support times between 5-15 minutes. I am firmly of the
opinin that 5 minutes is an absolute minimum and that times in excess
of 15 minutes are a gravy train. Except for the minimum, I am easily
persuaded, however.

response 49 of 61:

Apr 6 14:06 UTC 1996

Yes, of course. If you halt the machine, and then power comes back on,
before the UPS shutoff, there is nothing to actually reboot the machine,
and it sits at the monitor prompt until a human comes over and types in
a bootup command. In single user mode, you can still be running a program
that monitors the UPS, but the system is quiescent and the disks are synced.