[Oneiric-Topic] Simplifying system sleep functions

[Tony Espy]

On 04/22/2011 10:27 AM, Ted Gould wrote:
> On Thu, 2011-04-21 at 12:06 -0400, Tony Espy wrote:
>> On 04/21/2011 11:49 AM, Ted Gould wrote:
>>> On Thu, 2011-04-21 at 11:34 -0400, Tony Espy wrote:
>>>> On 04/19/2011 08:09 PM, Jason Warner wrote:
>>>>> Hi Everyone - Sending this on behalf of John Lea, desktop design lead.
>>>>>
>>>>> ==============================================
>>>>>
>>>>> Currently Ubuntu contains two separate sleep functions, suspend and
>>>>> hibernate.  This choice confuses users and is a un-necessary
>>>>> complication to 'sleeping' the computer.  The proposed change is to
>>>>> combine both 'suspend' and 'hibernate' into a single 'sleep' function.
>>>>>     When the user presses 'sleep', the computer should both suspend and
>>>>> hibernate simultaneously.  The computer remains suspended for a set
>>>>> period of time (e.g. 30min) or until the battery charge falls below a
>>>>> set level.  At the point the suspend state is discarded, and if  the
>>>>> user wakes the computer after this point their state is restored from
>>>>> hibernate.  However if the user wakes the computer before the suspend
>>>>> state is discarded, the computer is restored from 'suspend' and the
>>>>> 'hibernate' state is discarded.
>>>>
>>>> I'm not a fan of this idea.
>>>>
>>>> If suspend works for the vast majority of users, why complicate it by
>>>> adding a timed "auto-hibernate" to the equation?  As a few folks have
>>>> pointed out, what if hibernate fails?  What if the BIOS doesn't properly
>>>> support a wake timer?
>>>>
>>>> I'm pretty sure the latter criteria for triggering hibernate ( critical
>>>> low-battery event while suspended ) already works.  It essentially wakes
>>>> the system from suspend, the power manager notices the battery is
>>>> critically low, and invokes a hibernate.  The timed scenario would work
>>>> in a similar manner, except that after a timer event wakes the system,
>>>> the power manager would have to have added logic to trigger the hibernate.
>>>>
>>>> I'm much more in favor of hiding or even removing hibernate from the UI,
>>>> as long as it remains an option for "critical low-battery" event for
>>>> those systems that properly support hibernate.
>>>
>>> I think these are all valid cases, but I think that we should support
>>> this feature.  I think how we should handle this is with a whitelist if
>>> machines that we know hibernate works on.  We can provide instructions
>>> on adding your machine to that list if you want.  Otherwise machines
>>> that get certified by a vendor that cares about Ubuntu could ship their
>>> machine in that whitelist.
>>
>> Two words come to mind..."maintenance nightmare".  ;)
>>
>> After having lived through OEM-hell the last three months dealing with
>> ACPI stress testing and hibernate failures on Sandy Bridge machines, the
>> idea of maintaining a whitelist of machines that are known to have a
>> working hibernate function, doesn't seem very practical to me.
>
> I'm confused, wouldn't your work there be effectively building that
> whitelist?  Sounds like work you've already done ;-)

Ah, there you go, volunteering me for work items!  We can talk about 
this over beers in Budapest!

>>> What I think this does, and I don't believe it's really a bad thing, is
>>> makes it so there are effectively two Ubuntu experiences.  That which
>>> you get from installing off of the CD on random hardware, and that which
>>> you get when you use a hardware vendor that cares about Ubuntu.  I think
>>> that we need to make the experience the best we can for hardware vendors
>>> that want to participate in Ubuntu -- and provide reasonable fallback
>>> for those who don't.
>>
>> Personally, if we really want to consider this idea, I think we need to
>> put cycles into making hibernate work better first ( faster, more user
>> feedback, ... ).
>>
>> Another alternative would be to explore something more radical, along
>> the lines of what OS X does, which actually tries to combine hibernate
>> and sleep as opposed to running them in a serial fashion as proposed.
>
> So I guess that'd be the list of things we should discuss in the
> session.  What are the requirements and changes we'd need to make
> hibernate work well enough to make this a reality?  We can't budget time
> if we don't know what we want :-)
>
> Also, I thought this *was* how OSX did things.  Can you explain how that
> works as I don't know.

My understanding is that macbooks running the latest versions of OS X, 
always write a hibernate image to disk when sleeping, however the 
machine still goes into S3 (sleep).  If the system is resumed and power 
has never dropped below critical level, then it's just like a normal 
suspend/resume.  If however the system runs out of juice...the next time 
it's booted, it will resume from the hibernate image.

This is different than a normal Ubuntu install, where sleep == suspend, 
with no hibernate image written.  If the battery reaches a critical 
level, the machine is woken up, and the power manager will initiate a 
hibernate ( unless the user has changed the default setting ).

pm-utils does support something called pm-suspend-hybrid which acts like 
the OS X sleep, but I have little experience with it.

Does anybody care to comment on the current state of pm-suspend-hybrid?

Regards,
/tony




>
> 		--Ted
>
>