Auto-shutdown support for power outage - hint to Sun's Server Group ?

[Eric Ho]

This is just a suggestion -- perhaps somebody at Sun's Server Group might
want to at least start planning to add support to the kernel and in SunOS
in general so that it can interact with *temporary* power backup devices.
By temporary, I mean those boxes that will kick off automatically when the
voltage coming from the wall/floor socket drops below a certain threshold
and it'll supply power from its batteries until its own juices run out or
the power from the wall comes back up above the threshold.

I think that every future large servers sold by Sun should somehow
incorporate such temporary backup devices -- both to protect disk drives
and to warn other users & systems -- this is particulary important in a
large setup where you've lots of drives attached to the server.  Very
often, the power outage only lasts for a brief moment (as in a
thunderstorm for instance) and will come back up to normal -- and this
means that the disk drives will come online and offline like a yoyo -- not
too good for the drives even when they've autoparking arms.

The kernel/SunOS support mainly would log such events in /var/adm/messages
and do auto-shutdown if the voltage from the wall socket remains dead/low
for more than say 5~10 minutes -- thereby giving enough time for users &
other systems to finish off before its batteries run out.  The device also
should be able to send warning interrupts to the kernel when its batteries
are low -- e.g. it has "served" several brief power blackouts.  This
assumes the device is able to switch when the voltage coming from the wall
is back to normal again hence saving battery juice.  When the batteries
are low, sysadmin people should be notified and if they still don't change
batteries then the next blackout will be a permanent auto-shutdown no
matter how brief it is - hence guarantee that there will always be enough
batteries juice for any blackouts.

One can get those boxes where the batteries last for less than say 15
minutes thereby reducing cost.  Oh well, ... I wonder if anyone in the
Server Group might do something about this.

Eric Ho  
Princeton University
eho at clarity.princeton.edu