At the moment we set:
# CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE is not set CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE=y # CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND is not set # CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE is not set CONFIG_CPU_FREQ_GOV_PERFORMANCE=y CONFIG_CPU_FREQ_GOV_POWERSAVE=m CONFIG_CPU_FREQ_GOV_USERSPACE=y CONFIG_CPU_FREQ_GOV_ONDEMAND=m CONFIG_CPU_FREQ_GOV_CONSERVATIVE=m
This is not ideal from a power-saving point of view.
In an ideal world we would:
* remove CONFIG_CPU_FREQ_GOV_CONSERVATIVE -- ondemand does a better job on all workloads * remove CONFIG_CPU_FREQ_GOV_USERSPACE -- we have nothing in userspace that needs this sort of control, and if we did, the latency would be horrible * remove CONFIG_CPU_FREQ_GOV_POWERSAVE -- ondemand automatically throttles down to lowest, and is just a hardcoded state * compile into the kernel CONFIG_CPU_FREQ_GOV_ONDEMAND -- we really want to be running this on all systems that support it * set ONDEMAND or PERFORMANCE to default as USERSPACE is just changed to something else by cpuspeed. You really don't want to be using USERSPACE at all.
Matthew Garrett and I are working on a latency profile for power management, and having all these modules potentially loaded is bad.
Comments?
Richard.
On Fri, 27 Jun 2008 17:13:24 +0100 Richard Hughes hughsient@gmail.com wrote:
At the moment we set:
# CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE is not set CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE=y # CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND is not set # CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE is not set CONFIG_CPU_FREQ_GOV_PERFORMANCE=y CONFIG_CPU_FREQ_GOV_POWERSAVE=m CONFIG_CPU_FREQ_GOV_USERSPACE=y CONFIG_CPU_FREQ_GOV_ONDEMAND=m CONFIG_CPU_FREQ_GOV_CONSERVATIVE=m
This is not ideal from a power-saving point of view.
In an ideal world we would:
- remove CONFIG_CPU_FREQ_GOV_CONSERVATIVE -- ondemand does a better
job on all workloads
- remove CONFIG_CPU_FREQ_GOV_USERSPACE -- we have nothing in
userspace that needs this sort of control, and if we did, the latency would be horrible
- remove CONFIG_CPU_FREQ_GOV_POWERSAVE -- ondemand automatically
throttles down to lowest, and is just a hardcoded state
- compile into the kernel CONFIG_CPU_FREQ_GOV_ONDEMAND -- we really
want to be running this on all systems that support it
- set ONDEMAND or PERFORMANCE to default as USERSPACE is just
changed to something else by cpuspeed. You really don't want to be using USERSPACE at all.
Matthew Garrett and I are working on a latency profile for power management, and having all these modules potentially loaded is bad.
Comments?
I totally agree with your suggestions.
Richard Hughes wrote:
In an ideal world we would:
- compile into the kernel CONFIG_CPU_FREQ_GOV_ONDEMAND -- we really
want to be running this on all systems that support it
- set ONDEMAND or PERFORMANCE to default as USERSPACE is just changed
to something else by cpuspeed. You really don't want to be using USERSPACE at all.
How can an administrator set a known constant frequency, so that the CPU might be able to deliver the same amount of work per unit time, over a span of half an hour? Some performance measurement and tuning is much simpler when this is so.
On Fri, Jun 27, 2008 at 6:13 PM, Richard Hughes hughsient@gmail.com wrote:
You really don't want to be using USERSPACE at all.
seems like cpufreq-applet uses it....
On Fri, 2008-06-27 at 21:16 +0200, drago01 wrote:
On Fri, Jun 27, 2008 at 6:13 PM, Richard Hughes hughsient@gmail.com wrote:
You really don't want to be using USERSPACE at all.
seems like cpufreq-applet uses it....
Sure, it shouldn't. If you're using userspace for thermal or latency reasons, then a setuid applet is totally the wrong way to achieve both of these :-)
Maybe we can just use these as loadable modules (i.e. not built default) rather than built-in and loaded by default.
DaveJ, do these suggestions seem acceptable?
Richard.
On Fri, Jun 27, 2008 at 10:01 PM, Richard Hughes hughsient@gmail.com wrote:
On Fri, 2008-06-27 at 21:16 +0200, drago01 wrote:
On Fri, Jun 27, 2008 at 6:13 PM, Richard Hughes hughsient@gmail.com wrote:
You really don't want to be using USERSPACE at all.
seems like cpufreq-applet uses it....
Sure, it shouldn't. If you're using userspace for thermal or latency reasons, then a setuid applet is totally the wrong way to achieve both of these :-)
its not a setuid applet .. something seems to allow non root to do this (hal? consolekit? pam? udev? .. dunno)
On Fri, 2008-06-27 at 22:56 +0200, drago01 wrote:
On Fri, Jun 27, 2008 at 10:01 PM, Richard Hughes hughsient@gmail.com wrote:
On Fri, 2008-06-27 at 21:16 +0200, drago01 wrote:
On Fri, Jun 27, 2008 at 6:13 PM, Richard Hughes hughsient@gmail.com wrote:
You really don't want to be using USERSPACE at all.
seems like cpufreq-applet uses it....
Sure, it shouldn't. If you're using userspace for thermal or latency reasons, then a setuid applet is totally the wrong way to achieve both of these :-)
its not a setuid applet .. something seems to allow non root to do this (hal? consolekit? pam? udev? .. dunno)
It currently uses consolehelper to get root. IMO, it shouldn't allow setting frequencies at all.
On Fri, Jun 27, 2008 at 09:01:34PM +0100, Richard Hughes wrote:
On Fri, 2008-06-27 at 21:16 +0200, drago01 wrote:
On Fri, Jun 27, 2008 at 6:13 PM, Richard Hughes hughsient@gmail.com wrote:
You really don't want to be using USERSPACE at all.
seems like cpufreq-applet uses it....
Sure, it shouldn't. If you're using userspace for thermal or latency reasons, then a setuid applet is totally the wrong way to achieve both of these :-)
Maybe we can just use these as loadable modules (i.e. not built default) rather than built-in and loaded by default.
DaveJ, do these suggestions seem acceptable?
Having the userspace governor built-in means absolutely nothing in terms of overhead, until something in userspace actually uses it.
When the cpuspeed init script starts up, the first thing it does is check if the CPU is on the whitelist for using ondemand, and if so, it starts up ondemand. Not a single line of the userspace governor code gets run in this case.
The only time the above isn't true is when the CPU isn't on that whitelist, when it's incapable of running ondemand, in which case we need to use.. ta-da... userspace, and then we start the cpuspeed process.
Again, if you're seeing overhead from using userspace, it's due to your CPU being crap. There's nothing we can do about it. Whilst ondemand will load on some of these CPUs, the associated overhead of switching is very noticable on benchmarks.
Even 'conservative' was too demanding for some of the challenged CPUs.
'crap' here doesn't mean really old stuff too. Any pre-centrino Intel CPU, any VIA CPU before Nehemiah generation, all mobile Athlons.
We're using ondemand on all K8's too, but the first generation also sucked iirc, but we're just sucking it up because a) it makes the already convoluted startup script even more messy and b) no-one can remember which stepping/models were affected.
Dave
On Fri, Jun 27, 2008 at 05:13:24PM +0100, Richard Hughes wrote:
At the moment we set:
# CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE is not set CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE=y # CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND is not set # CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE is not set CONFIG_CPU_FREQ_GOV_PERFORMANCE=y CONFIG_CPU_FREQ_GOV_POWERSAVE=m CONFIG_CPU_FREQ_GOV_USERSPACE=y CONFIG_CPU_FREQ_GOV_ONDEMAND=m CONFIG_CPU_FREQ_GOV_CONSERVATIVE=m
This is not ideal from a power-saving point of view.
In an ideal world we would:
- remove CONFIG_CPU_FREQ_GOV_CONSERVATIVE -- ondemand does a better job
on all workloads
- remove CONFIG_CPU_FREQ_GOV_USERSPACE -- we have nothing in userspace
that needs this sort of control, and if we did, the latency would be horrible
needed for the cpuspeed governor
- remove CONFIG_CPU_FREQ_GOV_POWERSAVE -- ondemand automatically
throttles down to lowest, and is just a hardcoded state
- compile into the kernel CONFIG_CPU_FREQ_GOV_ONDEMAND -- we really
want to be running this on all systems that support it
- set ONDEMAND or PERFORMANCE to default as USERSPACE is just changed
to something else by cpuspeed. You really don't want to be using USERSPACE at all.
Not all CPUs are capable of running ondemand because of the latency they incur during transitions.
Matthew Garrett and I are working on a latency profile for power management, and having all these modules potentially loaded is bad.
I don't follow this. Can you show whatever numbers you have that you're basing this on ?
Dave
On Fri, 2008-06-27 at 15:18 -0400, Dave Jones wrote:
On Fri, Jun 27, 2008 at 05:13:24PM +0100, Richard Hughes wrote:
At the moment we set:
# CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE is not set CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE=y # CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND is not set # CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE is not set CONFIG_CPU_FREQ_GOV_PERFORMANCE=y CONFIG_CPU_FREQ_GOV_POWERSAVE=m CONFIG_CPU_FREQ_GOV_USERSPACE=y CONFIG_CPU_FREQ_GOV_ONDEMAND=m CONFIG_CPU_FREQ_GOV_CONSERVATIVE=m
This is not ideal from a power-saving point of view.
In an ideal world we would:
- remove CONFIG_CPU_FREQ_GOV_CONSERVATIVE -- ondemand does a better job
on all workloads
- remove CONFIG_CPU_FREQ_GOV_USERSPACE -- we have nothing in userspace
that needs this sort of control, and if we did, the latency would be horrible
needed for the cpuspeed governor
Sure, I might have been a bit hasty in saying _remove_ - I don't really care about old CPU types but I appreciate lots of people do.
- remove CONFIG_CPU_FREQ_GOV_POWERSAVE -- ondemand automatically
throttles down to lowest, and is just a hardcoded state
- compile into the kernel CONFIG_CPU_FREQ_GOV_ONDEMAND -- we really
want to be running this on all systems that support it
- set ONDEMAND or PERFORMANCE to default as USERSPACE is just changed
to something else by cpuspeed. You really don't want to be using USERSPACE at all.
Not all CPUs are capable of running ondemand because of the latency they incur during transitions.
Right, isn't the latency exported by the kernel? Surely it's a userspace decision (policy) if the latency is acceptable? The little tool Matthew and I prototyped yesterday allows an admin to tweak the latency they want in the whole system, and depending on the latency acceptable, different things like ALPM, ASPM and the governor are put in different modes. Obviously these new switches help in powersaving (lots) but each add appreciable latency which may be unacceptable.
Matthew Garrett and I are working on a latency profile for power management, and having all these modules potentially loaded is bad.
I don't follow this. Can you show whatever numbers you have that you're basing this on ?
Well, not bad from a speed point of view, bad from a complexity view. If we compile in ondemand, performance and userspace then we can leave it up to policy in the system to decide what makes sense. The policy in userspace can get the latency of the ondemand change and make a decision on that.
Some of the logic in cpuspeed seems a little complex, and I'm wondering if we need to be running a service like cpuspeed on a desktop laptop.
Richard.
On Fri, Jun 27, 2008 at 05:13:24PM +0100, Richard Hughes wrote:
- remove CONFIG_CPU_FREQ_GOV_POWERSAVE -- ondemand automatically
throttles down to lowest, and is just a hardcoded state
I don't think removal of powersave governor is good idea. Generally ondemand governor does great job but in some cases doesn't. For example when I play some films in mplayer ondemand sets frequency to max which is not needed, of course.
Powersave governor is also good in case that you have bad fan in your laptop and you are going to compile some big source. Without powersave it is not possible (yes, it really happens :) )
Matthew Garrett and I are working on a latency profile for power management, and having all these modules potentially loaded is bad.
Comments?
I think we should preserve ondemand and powersave governors (and potentialy others as Dave Jones wrote in this thread). Please don't drop them in favour of your project which might be generally better but I believe there are cases where current governors are better.
Adam
On Mon, 2008-06-30 at 09:10 +0200, Adam Tkac wrote:
On Fri, Jun 27, 2008 at 05:13:24PM +0100, Richard Hughes wrote:
- remove CONFIG_CPU_FREQ_GOV_POWERSAVE -- ondemand automatically
throttles down to lowest, and is just a hardcoded state
I don't think removal of powersave governor is good idea. Generally ondemand governor does great job but in some cases doesn't. For example when I play some films in mplayer ondemand sets frequency to max which is not needed, of course.
Right, so we need to fix ondemand to be cleverer.
Powersave governor is also good in case that you have bad fan in your laptop and you are going to compile some big source. Without powersave it is not possible (yes, it really happens :) )
Right, thermal management is similar to power management for the action but not for the policy. I don't think forcing the lowest speed setting is the correct way to fix this. If the laptop is running cool, why use the slowest speed?
Matthew Garrett and I are working on a latency profile for power management, and having all these modules potentially loaded is bad.
Comments?
I think we should preserve ondemand and powersave governors (and potentialy others as Dave Jones wrote in this thread). Please don't drop them in favour of your project which might be generally better but I believe there are cases where current governors are better.
Right, cheers for your feedback. In view of everybodies comments, what about the following:
* Compile _into_ the kernel ondemand, performance, powersave and userspace. * Default to performance in the kernel rather than userspace * Build as a module conservative with the view of just fixing ondemand if there are any special use-cases that conservative is better at * Export the P and C state latency to userspace and let the system policy dictate the governor. For instance, even for machines that have a long latency for changing P states should be able to use ondemand if we want to save maximum power.
How does that sound?
Richard.
On Monday 30 June 2008 05:54:32 am Richard Hughes wrote:
Right, cheers for your feedback. In view of everybodies comments, what about the following:
- Compile _into_ the kernel ondemand, performance, powersave and
userspace.
Sounds reasonable.
- Default to performance in the kernel rather than userspace
What's the difference? Both leave the cpu at its max speed all the time, unless the cpuspeed daemon gets started up in the userspace case.
- Build as a module conservative with the view of just fixing ondemand
if there are any special use-cases that conservative is better at
- Export the P and C state latency to userspace and let the system
policy dictate the governor. For instance, even for machines that have a long latency for changing P states should be able to use ondemand if we want to save maximum power.
How does that sound?
Mostly sane. System policy dictating governor over the ugliness we do in the cpuspeed init script would be nice. Even nicer would be if we could outright get rid of the initscript (not sure what people who need the cpuspeed daemon are to do in that case though).
On Mon, Jun 30, 2008 at 09:10:28AM +0200, Adam Tkac wrote:
On Fri, Jun 27, 2008 at 05:13:24PM +0100, Richard Hughes wrote:
- remove CONFIG_CPU_FREQ_GOV_POWERSAVE -- ondemand automatically
throttles down to lowest, and is just a hardcoded state
I don't think removal of powersave governor is good idea. Generally ondemand governor does great job but in some cases doesn't. For example when I play some films in mplayer ondemand sets frequency to max which is not needed, of course.
The same can be achieved by altering /sys/devices/system/cpu/*/cpufreq/scaling_max_freq, but it's still likely that you're consuming less power when ondemand is setting your frequency to max. An idle fast processor consumes less power than an active slow one.
Powersave governor is also good in case that you have bad fan in your laptop and you are going to compile some big source. Without powersave it is not possible (yes, it really happens :) )
http://lkml.org/lkml/2008/6/16/100
I think we should preserve ondemand and powersave governors (and potentialy others as Dave Jones wrote in this thread). Please don't drop them in favour of your project which might be generally better but I believe there are cases where current governors are better.
I'm open to indications as to what these are :) Powersave is semantically identical to ondemand with scaling_max_freq altered. Performance is semantically identical to ondemand with scaling_min_freq altered.
kernel@lists.fedoraproject.org
-
Adam Tkac -
Arjan van de Ven -
Bastien Nocera -
Dave Jones -
drago01 -
Jarod Wilson -
John Reiser -
Matthew Garrett -
Richard Hughes