On Mon, 17 Apr 2017 15:45:49 +0000, Frank Miles wrote:

> On Sun, 16 Apr 2017 19:42:50 +0000, Gordon Henderson wrote:
>
>> In article ,
>> Frank Miles   wrote:
>>>On Fri, 14 Apr 2017 19:25:11 +0100, Ahem A Rivet's Shot wrote:
>>>
>>>> On Fri, 14 Apr 2017 16:30:48 -0000 (UTC)
>>>> Frank Miles  wrote:
>>>>
>>>>> I have an embedded application for the RPi-3.  Unfortunately it's likely
>>>>> that uses will remove power without properly shutting the RPi down.
>>>>> Therefore I've got a 'power-going-down' bit of hardware which sets a GPIO
>>>>> bit causing a proper shutdown.  The question, of course, is how long does
>>>>> a shutdown take?
>>>>
>>>>  It depends - mostly on how fast the SD card is, how many dirty
>>>> buffers need to be flushed and how long processes take to die. Best bet is
>>>> to measure it under your worst case loads with the card you intend to use
>>>> and add a healthy safety factor.
>>>
>>>Ok, reframing :
>>>  What's the easiest way to detect when shutdown is complete?
>>
>> Some boards use the serial Tx line. This is set low when shutdown is
complete.
>
> I'll take a look at that too once I get my tech to make me the power cable
> with separable conductors needed for monitoring the current consumed.  I'm
> also curious to see how much current the RPi consumes once it is shut-down
> (since it reboots after some 10s or so it must be running, at a minimum, some
> kind of timer/clock).
>
>> Other strategies are to minimise SD card writes during normal use, but we
>> don't know what the application is.. However stuff I do is things like
>> remove logging (like rsyslogd - no idea how to stop systemd writing to
>> SD card, I don't use it), mounting /tmp/ as ramdisk (see /etc/defaults/shm )
>> and generally tuning application to minimise writes, if possible.
>>
>> Gordon
>
> It's running a GUI that controls a research instrument.  The only disk
activity
> is very infrequent, to save altered instrument configuration/settings.  At
this
> point (the users may come up with changes in the future), there's no data
> logging.
>
> Excellent idea to kill any syslog activities, etc., thanks!
>
>     -F

I've tried a few experiments to reduce the shutdown/poweroff demand, to little
effect.  These all used "poweroff" or equivalent while the power supply
actually remained on.

My test unit (connected to a 7" touch screen, powered separately) draws:
  ~ 350mA while idle (with or without X)
  ~ 500mA (avg) during the ~2sec it seems to take to shut down
  ~ 100mA after shutdown appears complete
The 2s of "extended current draw" occurs after the screen blanks,
which is nearly instantaneous - i.e. the screen blanking is
misleadingly fast.

Preliminary attempts to quicken or reduce the magnitude of current
during shutdown - by not using X, or stopping rsyslog have been
completely ineffective.

The Tx line drops as suggested; at least it's coincident with the majority
of the current dropping from ~500mA to ~100mA (there's 0.4 to 0.5s of
75mA more after the Tx line drop - hopefully that's not more "disk"
activity).

Yuck.  Probably gonna have to beef up my 5V supply's stored energy to
avoid system flash data corruption.  I may yet try a few things - there's
lots of activity posted to syslog.

   -F

--- SoupGate-Win32 v1.05