UPS (power-backup) boxes for Sun servers ??

[Paul Evan Matz]

In article <9852 at brazos.Rice.edu> eho at clarity.princeton.edu (Eric Ho) writes:
>
>Has anyone out there been using those UPS (Uninterruptible power system)
>boxes for Sun servers ?  In particular, I want to get opinions about the
>UPS boxes from Elgar (in San Diego)...

So far, the only UPS we've found that will work well with SunOS is the IPS
400 + 600 from Elgar (800-854-2213).  You can buy "Unisafe" for SunOS
($99) which is a set of sh scripts that monitor the state of the UPS via a
serial port, and shutdown the Sun in the case of a power failure.  The
price of the UPS;  something like $500-600.

One unique feature that the Elgar has is the ability to shutoff its UPS
output power to the Sun once it has halted.  This is very handy when the
power failures are frequent or long.  (The key here is that the UPS has a
built-in delay circuit which waits for 20 seconds after being signaled
before it shuts off its output power;  The Sun signals, then halts.  This
gives it 20 seconds to halt and flush.  Without this feature, it is
difficult to make sure the disk is flushed while under software control.
Once the Sun has halted, it can't signal the UPS to shutoff.  Your basic
catch-22).

Powering a monitor is problematic for most UPS, this one included.  The
problem is not power usage, by peak current.  A 16" color monitor might
use as much as 3 Amps peak (although only about 100 Watts average); That
coupled with the Sun will amount to more then the max. for most UPSs
(3.3amps max).  Therefore, don't plan on powering you monitor.  This will
usually not be a problem for servers.

Power just a SPARCStation 1 w/ 2 104M disks, the projected power backup
time is something like 1.5 hours.  I ran it for about 20 minutes without
any noticable battery discharge (as shown by a battery capacity bar graph
display on the UPS's front panel).

The UPS monitor daemon process senses when power goes down by being
blocked on a read of the tty;  The UPS asserts CD when power goes down,
thereby unblocking the daemon, who proceeds to start a countdown.
(There's no connection to the Receive data pin).  There are various
configurable parameters that define how long the daemon ignores a power
outage, how long it waits after it notifies users to logout and before it
starts to kill off processes, etc.  The daemon is built out of scripts, so
it is easy to checkout and modify if you need to, although I think they
could use a little more work.

When it is finally time to shut off the UPS, the script signals the UPS by
writing a file of all zeros out to the tty (the Transit Data pin is
connected to the UPS).  Then, unless the AC power comes back on, the UPS
shuts off power to the Sun in 20 seconds.  The script hammers as many
processes as it can (including itself, by the way), sleeps for 20 seconds
and then reboots (so if the power does come back on, the machine won't be
sitting there halted).  There is a race condition where, depending on the
spacing of multiple power outages, you can get your power shut off during
a reboot.  (This is one of the things that need to be fixed in that daemon
script).

Another problem with the Elgar software is its installation instructions.
On a SPARCstation, there are a few details which are omitted, like setting
carrier-detect-ignore to false, and adding the serial port to gettytab.  I
think they are working on fixing this, though.

I have no real info on reliability.  So far, so good.

The support from Elgar (Tod Fox) is very good.

Hope this is of interest.