Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 05 May 2014 09:53:44 +1000
From:      Dewayne Geraghty <dewayne.geraghty@heuristicsystems.com.au>
To:        freebsd-hackers@freebsd.org
Subject:   Re: Leaving the Desktop Market
Message-ID:  <5366D308.5090808@heuristicsystems.com.au>
In-Reply-To: <53661C77.5030806@freebsd.org>
References:  <CAF6rxgkeBozvfV-L0%2BrFZ6fWRn0=Gi3BNq1kPL=-HTq0TD6MkQ@mail.gmail.com> <A70900DF-4BAA-427F-8731-01211FFD1887@mail.turbofuzz.com> <3F7430D7-3C0F-43E1-8EBD-8AA4F701497C@FreeBSD.org> <20140503155745.GA2457@La-Habana> <CAJ-VmokHQvb2S-1c08J3RECUMosQ%2Bcd1gdB91LAb5famfKqP5w@mail.gmail.com> <20140503192305.GA1847@La-Habana> <CAJ-Vmon1ZYX18SN1f=z9N=hSgpZrAxXMxiJBUDWk6LYvMRiHmw@mail.gmail.com> <CAN6yY1uo08PDsYJ2KQNjw4hYuNsNJUvRu1b0cG%2B387F6Owkukg@mail.gmail.com> <536592D1.7080403@freebsd.org> <CAN6yY1tdsitXmGeu87OBw6RZ_s0qDvuoYXbTJVjnr74k6oM4Ag@mail.gmail.com> <CAJ-Vmo=zXyuyARapz4UCaTwMtWtW_8ys0pgssHOOUhnA-5Bt4g@mail.gmail.com> <CAN6yY1ukYxuf=jwBVbepCg84k54HdquyCiwpRx%2BB=SQ1hb0efg@mail.gmail.com> <CAJ-Vmokq-e7qMNZq_XVh3XBidAg7o6y4MOscvJYXSZB=dN9EjA@mail.gmail.com> <20140504112807.1136d108@x23> <53661C77.5030806@freebsd.org>

next in thread | previous in thread | raw e-mail | index | archive | help

On 4/05/2014 8:54 PM, Stefan Esser wrote:
> Am 04.05.2014 11:28, schrieb Marko Zec:
>> On Sun, 4 May 2014 01:13:48 -0700
>> Adrian Chadd <adrian@freebsd.org> wrote:
>>
>>> [snip]
>>>
>>> ok, how about this to start with:
>>>
>>> http://people.freebsd.org/~adrian/power/20140504-powerd-1.diff
>>>
>>> This defaults powerd to set minfreq to 50% of the hardware available
>>> maxfreq, unless minfreq is explicitly given on the command line.
>> As already noted earlier in this thread, if you disable
>>
>> hint.p4tcc.0.disabled=1
>> hint.acpi_throttle.0.disabled=1
>>
>> then the kernel does not even expose expose those silly minimum
>> "frequencies", and your problem goes away without patching powerd.
>>
>> A more reasonable and simpler patch would be to disable the two
>> offending throttling drivers by default, I really cannot see a single
>> reason why do we need them at all, less why they are enabled.
>>
>> Marko
> Very true and this topic has come up so many times during for at least a
> decade (if not much longer). Throttling had its use at a time, but
> this time is long gone.
>
> Throttling is enabled unless probing of the "p4tcc" and "acpi_throttle"
> pseudo-devices is disabled by device hints.
>
> As a first step, these hints could be set in /boot/defaults/loader.conf.
> They could still be overridden in /boot/loader.conf, if they really are
> required (anybody still got P4 systems running -CURRENT?).
>
> A better fix would be to disable throttling depending on the CPU model
> identified (or on the presence of any other power management solution).
>
> Since the CPUs are probed before any device, adding a "disable" hint
> for both throttling methods would be kind of a hack, but should work.
>
> If there is an agreement, that we should move into that direction, I'll
> try to create a patch for review.
>
> Regards, STefan
> _______________________________________________
> freebsd-hackers@freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-hackers
> To unsubscribe, send any mail to "freebsd-hackers-unsubscribe@freebsd.org"
>
>

Its really pleasing to see this discussion, the insightful historical
context from Kevin and Adrian's generous offer to take up the mantel, re
the proposal of the work required.

We've been using low powered servers since 2003.  Some of our constraints:
- at a couple of customer sites, air-conditioning is turned off after
18:00 so thermal controls are significant during summer
- medium businesses with tight budgets want to reduce electrical costs
(30 devices isn't uncommon)
- we're not running current on customer P4 machines but we are running
9.2 Stable on most. Working to eliminating P4tcc and acpi_throttling
seems to be beneficial so  disabling by default is a good idea for most,
and the requirement to enabling on the older systems should be
acceptable to most.  (As an FYI, the VIA chipsets, which we purchased 8
months ago only have C1 state  dev.cpu.0.cx_supported=C1/0 )
- sometimes its overlooked, that folks in less developed countries don't
have access to later generation equipment; and in my experience they
don't raise their financial situation, they quietly do without.   Not
our situation (btw), longevity & lower power consumption is a frequent
management choice.

And any help in (not) tuning powerd for each core/thread solution would
be appreciated.
An e.g night-time use at a thermally hot site with C1 state only, 
/usr/sbin/powerd -i 96 -r 80 -p 500 -n hadp for single core, and we peg
the low end at debug.cpufreq.lowest=800.  It's a black art, and may be
unavoidable :)

Regards, Dewayne



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5366D308.5090808>