SUMMARY:cut power on sun server

From: Andy J. Stefancik 206-234-3049 (ajs6143@eerpf001.boeing.com)
Date: Thu Jul 23 1992 - 06:54:54 CDT


        The general consensus from experience is that
cutting main power to the sun servers causes no problems.
        However, from an expertise standpoint, several people
felt, with good reason, that disks needed to be power down
first, then cpu. The opposite of I what I said. I can't really
tell if these responses mean that I shouldn't cut main power. In
my defense, all I can say is that I've been doing it cpu first
down for years, as have some other people I know, The simple reason
being that the fragile cpu board is more likely to get zapped by a
power spike from the disks.

Here is a list of the responses following my original query.

-------------------------------------------------------

>we're setting up a UPS monitor daemon for auto-shutdown, which
>is fine, but, we're also looking at adding a delay circuit for auto
>powerdown. We've been told by our facilities people that draining the
>UPS batteries is not good. The UPS is an 80KVA Exide. The sun's are
>670MPs with a combination of scsi and IPI drives.
>My question is this:
>What are the hardware effects of cutting main power to a Sun server
>as opposed to an orderly powerdown, i.e. - cpu first, then peripherals.
>Note that the server is "halted" when this occurs. Any insight or
>experience is welcome.
----------------------------------------------------------

From: bcsaic!ogicse!bit!jayl (Jay Lessert)

For what it's worth, I've done exactly this to all our servers (a mix
of Sun and Solbourne) at least once a year for the last 5 years, when
we do our annual Halon system check. I leave all the systems halted
but powered up, then have the maintenance guy trip the sensor. Power
to the whole room is cut. No ill effects. To be perfectly frank, it
never occured to me that there *could* be any harm caused, except the
obvious possibility of a power supply failure *any* time you cycle
power. So go for it.

----------------------------------------------------------

From: mdl@cypress.com (J. Matt Landrum)

I thought the only problem with draining the UPS batteries is that you
might get low power to the Suns. We halt our Suns about 5 or 10 minutes
before the battery power runs out then send a cutoff command to the UPS.
I have had no problems cutting off all the power at once like this, but
would be interested in hearing the results of your inquiry. What do the
people who sold you the UPS say about draining the batteries. I've heard
it can actually be good to totally drain rechargeable batteries (like
the ones in a calculator), but I'm sure the Sun wouldn't like it.

>Our UPS has lead acid batteries. Our facilities people are the
>ones telling us that draining the UPS batteries is not good. I know
>with nicads it is good.

---------------------------------------------------------------

From: geertj@ica.philips.nl

The 'delayed power-on' as found on 4/490's (guess it's on 670's too,
can't check because of budget cuts) can use a steering voltage to
switch things on/off. I guess you contact the firm that built those,
the address is on the module (the copper-coloured power switch with
'local-remote switch), and the 'remote' possibility can do what
you want.

>Don't know if this is referring to setting up the delay power-off
>after shutdown, or how the sun reacts when main power is cut.

-------------------------------------------------------------

From: kalli!kevin@fourx.Aus.Sun.COM (Kevin Sheehan {Consulting Poster Child})

If it is halted, no problem. Building a system that can't handle a power
failure is pretty silly. The 600's have a sequencing power supply, so
coming back up shouldn't be a problem either.

>sigh, some of us still remember the old days. Happy that this is
>the case today though.

---------------------------------------------------------------

>From kean@argh.ucs.orst.edu Thu Jul 16 21:47:06 1992

What power rating did you use for the 670's when rating the UPS? I have a
690 I want to put on a UPS and the Sun literature isn't clear on the power
consumption of the VME cardcage and the disks and such.

>Again, our facilities people figured this out. We are currently at
>25% load as it was designed for expansion. At 80KVA. its a big one.
>that handle 6 4/470s (soon to go to 670mps) and 2 epochs.

---------------------------------------------------------------

From: exudnw@exu.ericsson.se (Dave Williams)

I think that the disks should be powered down *first*, then the CPU (this
assumes the CPU is halted). I have seen CPU's cause disk problems when
the disks were still up and the CPU power was cycled.
 
This may be comp.sun.folklore, but it's what I'd do (I have 400 Sun clients
& 13 490/690's on a *huge* UPS shared with an IBM 3090-400).

---------------------------------------------------------------

From: wolfgang%sunspot.nosc.mil@nosc.mil

We have never seen a problem caused by complete power
removal from a computer system. Where we have seen problems is
when power is applied, then removed and applied again after a short
duration. Someone toggling the circuit breaker would cause this,
but we have also seen it happen on shipboard applications and during
storms on land.
 
        We have also heard from UPS manufacturers that cycling
the batteries shortens their lifetimes, but after all, that is
what they are there for. We just don't discharge them on purpose,
only test them for a minute or so at load.

----------------------------------------------------------------

From: Mike Raffety <miker@sbcoc.com>

You have a VERY small chance of the following:
 
1. Power begins to drop.
2. CPU puts some random garbage out on the disk controller as a result
of the low voltage.
3. Disk manages to perform the garbage, which happens to be a write,
before it loses all power.
 
VERY small chance. BTW, you are supposed to power off disks FIRST, then
CPU. And power up CPU first, then disks, for just this reason. It looks
like you have it reversed ("cpu first, then peripherals").
 
>This may be true, I guess I got this from pdp11s where the peripherals
>had to on on first in order to boot. However I may have a memory
>problem, but like they say around here, "that's the second thing
>to go".

Andy Stefancik Internet: as6143@eerpf001.ca.boeing.com
Boeing Commercial Airplane G. UUCP: ...!uunet!bcstec!eerpf001!as6143
P.O. Box 3707 MS 64-25 Phone: (206) 234-3049
Seattle, WA 98124-2207



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:45 CDT