Re: power measurement

Re: power measurement

[Andy Green]

On 1 August 2013 14:26, Comaschi, F. <fcomaschi@tue.nl> wrote:

Somehow this just appeared today.

> Dear Mike,
>
> yes I have seen Andy's presentation about using AEP, and I had the impression that some measurements would have been possible on my board. Could you tell me which is the exact reason why it is more advisable to use a NI DAQ rather than a AEP? I thought that an AEP would have integrated better with the ARM development chain of DS5.
>
> Also, could you please tell me what do you exactly mean by "Solder sense
> resistors across exposed capacitors"?

I think Mike must have meant inductors; since the capacitors are
between the rail and 0V adding the shunts there will result in a
direct short across the rail hopefully triggering something's
overcurrent detection or if not "letting the smoke out" of your
low-value shunt resistor.

(Also /Gross/Green/ ^^)

> Does it mean inserting a shunt resistors in series to an existing capacitor on the board? Would you or somebody else be able to suggest me any of such capacitors on the Arndale board?
> According to Andy' s presentation, I thought that the best way to do power measurement was rather looking for inductors on the input side of power regulators, and replace those ones with shunt resistors.

That is correct.

> Looking at the enclosed schematic, Andy told me that unfortunately it seems like the Arndale 5250 does not present any suitable spot where to perform accurate power measurements.

I couldn't see anything useful to measure ARM Vcore, which is ironic
since there are so many other rails with shunt resistors already on
that board.

-Andy

> Could you please tell me more about this?
>
> Thanks again for all the provided information.
>
> Best regards,
>
> Francesco
>
> ________________________________________
> From: Mike Turquette [mike.turquette@linaro.org]
> Sent: Wednesday, July 31, 2013 5:58 AM
> To: Comaschi, F.
> Subject: Re: power measurement
>
> Quoting Comaschi, F. (2013-07-10 05:49:36)
>> Dear Mike,
>>
>> I am Francesco Comaschi, a researcher at Eindhoven University of Technology. I am sending you this email because my group is interested in purchasing an ARM-powered board where to measure power consumption with the highest accuracy possible.
>> Ideally, the board we are looking for has the following characteristics:
>>
>> -) options for power-saving strategies (DVFS and power gating);
>> -) options for measuring power consumption on the board (possibly via software, maybe using lm_sensors), or via hardware otherwise (maybe using ARM Energy Probe?);
>> -) support available: since I do not have much experience, i would rather choose an option where either some work has already been done in terms of power management/power measurements, or some sort of support would be guaranteed (for example, regarding the Arndale board I could not get to work the DVFS through Linaro, and so far I didn'get any answer/feed-back from In Signal).
>>
>> Also, if you have any documentation/tutorial/reference to suggest me in order to understand how to properly perform DVFS/power gating on one of the boards supported by Linaro, I would be really grateful.
>> I was considering, among the others, the Arndale dual-board (that we already have), the Origen quad-core board and the Panda board. Also the Versatile Express family from ARM is very attractive of course, but terribly expensive.
>
> Hi Francesco,
>
> Arndale and Panda are both good options which have some support upstream
> for DVFS and power gating.
>
>> I have read about the possibility of using the AEP to measure energy on device:
>> http://www.linaro.org/documents/download/3f44bb2b53ebd7ef498202d496c8cadd5093c0a862c4a
>> and I was wandering if anything similar has been already done on the Origen and the Arndale.
>
> AEP is very limiting. Did you see Andy Gross' presentation on using the
> AEP? It might be good for observing some DVFS transitions, but is
> essentially useless for high-accuracy low-power idle transitions.
>
> It is important to know that measuring high-performance workloads is not
> the same as measuing low power workloads. You might choose different
> resistor values depending on whether you are interested in measuring
> active frequency scaling versus idle power gating.
>
> There is still no easy way to do this. I suggest purchasing a National
> Instruments DAQ, one of the USB 6000 series should be good. Solder sense
> resistors across exposed capacitors on the boards and measure that way.
> It is how most of us do it when we want to get serious about measuring
> power.
>
> If you have further questions please Cc the Linux ARM kernel mailing
> list. It means more people can answer your questions and future
> researchers can benefit from your correspondence.
>
> Regards,
> Mike
>
>>
>> Also, from this page:
>> https://wiki.linaro.org/WorkingGroups/PowerManagement/Matrix1105
>> is is not clear to me if the cpu_idle and cpu_freq frameworks (which I think are used for power gating and DVFS respectively) are up and running on the mentioned boards, and if it would be possible to manage DVFS and power-gating directly from inside my application.
>>
>> Sorry for all the questions, but any help would be very much appreciated.
>>
>> Thanks in advance for your kind attention,
>>
>> Best regards,
>>
>> Francesco Comaschi
> _______________________________________________
> linaro-kernel mailing list
> linaro-kernel@lists.linaro.org
> http://lists.linaro.org/mailman/listinfo/linaro-kernel
>

Re: power measurement

[Viresh Kumar]

On 5 August 2013 14:02, Andy Green <andy.green@linaro.org> wrote:
> On 1 August 2013 14:26, Comaschi, F. <fcomaschi@tue.nl> wrote:
>
> Somehow this just appeared today.

Because it was stuck for Admin approval:

Reason:  Message body is too big: 260075 bytes with a limit of 100 KB

I just approved it today. :)

Re: power measurement

[Mike Turquette]

Quoting Andy Green (2013-08-05 01:32:41)
> On 1 August 2013 14:26, Comaschi, F. <fcomaschi@tue.nl> wrote:
> 
> Somehow this just appeared today.
> 
> > Dear Mike,
> >
> > yes I have seen Andy's presentation about using AEP, and I had the impression that some measurements would have been possible on my board. Could you tell me which is the exact reason why it is more advisable to use a NI DAQ rather than a AEP? I thought that an AEP would have integrated better with the ARM development chain of DS5.
> >
> > Also, could you please tell me what do you exactly mean by "Solder sense
> > resistors across exposed capacitors"?
> 
> I think Mike must have meant inductors; since the capacitors are
> between the rail and 0V adding the shunts there will result in a
> direct short across the rail hopefully triggering something's
> overcurrent detection or if not "letting the smoke out" of your
> low-value shunt resistor.
> 
> (Also /Gross/Green/ ^^)

Two embarrassing mistakes in one email. Yes I meant inductors and not
caps. Please leave the caps alone. Thanks for the corrections Andy.

Regards,
Mike

> 
> > Does it mean inserting a shunt resistors in series to an existing capacitor on the board? Would you or somebody else be able to suggest me any of such capacitors on the Arndale board?
> > According to Andy' s presentation, I thought that the best way to do power measurement was rather looking for inductors on the input side of power regulators, and replace those ones with shunt resistors.
> 
> That is correct.
> 
> > Looking at the enclosed schematic, Andy told me that unfortunately it seems like the Arndale 5250 does not present any suitable spot where to perform accurate power measurements.
> 
> I couldn't see anything useful to measure ARM Vcore, which is ironic
> since there are so many other rails with shunt resistors already on
> that board.
> 
> -Andy
> 
> > Could you please tell me more about this?
> >
> > Thanks again for all the provided information.
> >
> > Best regards,
> >
> > Francesco
> >
> > ________________________________________
> > From: Mike Turquette [mike.turquette@linaro.org]
> > Sent: Wednesday, July 31, 2013 5:58 AM
> > To: Comaschi, F.
> > Subject: Re: power measurement
> >
> > Quoting Comaschi, F. (2013-07-10 05:49:36)
> >> Dear Mike,
> >>
> >> I am Francesco Comaschi, a researcher at Eindhoven University of Technology. I am sending you this email because my group is interested in purchasing an ARM-powered board where to measure power consumption with the highest accuracy possible.
> >> Ideally, the board we are looking for has the following characteristics:
> >>
> >> -) options for power-saving strategies (DVFS and power gating);
> >> -) options for measuring power consumption on the board (possibly via software, maybe using lm_sensors), or via hardware otherwise (maybe using ARM Energy Probe?);
> >> -) support available: since I do not have much experience, i would rather choose an option where either some work has already been done in terms of power management/power measurements, or some sort of support would be guaranteed (for example, regarding the Arndale board I could not get to work the DVFS through Linaro, and so far I didn'get any answer/feed-back from In Signal).
> >>
> >> Also, if you have any documentation/tutorial/reference to suggest me in order to understand how to properly perform DVFS/power gating on one of the boards supported by Linaro, I would be really grateful.
> >> I was considering, among the others, the Arndale dual-board (that we already have), the Origen quad-core board and the Panda board. Also the Versatile Express family from ARM is very attractive of course, but terribly expensive.
> >
> > Hi Francesco,
> >
> > Arndale and Panda are both good options which have some support upstream
> > for DVFS and power gating.
> >
> >> I have read about the possibility of using the AEP to measure energy on device:
> >> http://www.linaro.org/documents/download/3f44bb2b53ebd7ef498202d496c8cadd5093c0a862c4a
> >> and I was wandering if anything similar has been already done on the Origen and the Arndale.
> >
> > AEP is very limiting. Did you see Andy Gross' presentation on using the
> > AEP? It might be good for observing some DVFS transitions, but is
> > essentially useless for high-accuracy low-power idle transitions.
> >
> > It is important to know that measuring high-performance workloads is not
> > the same as measuing low power workloads. You might choose different
> > resistor values depending on whether you are interested in measuring
> > active frequency scaling versus idle power gating.
> >
> > There is still no easy way to do this. I suggest purchasing a National
> > Instruments DAQ, one of the USB 6000 series should be good. Solder sense
> > resistors across exposed capacitors on the boards and measure that way.
> > It is how most of us do it when we want to get serious about measuring
> > power.
> >
> > If you have further questions please Cc the Linux ARM kernel mailing
> > list. It means more people can answer your questions and future
> > researchers can benefit from your correspondence.
> >
> > Regards,
> > Mike
> >
> >>
> >> Also, from this page:
> >> https://wiki.linaro.org/WorkingGroups/PowerManagement/Matrix1105
> >> is is not clear to me if the cpu_idle and cpu_freq frameworks (which I think are used for power gating and DVFS respectively) are up and running on the mentioned boards, and if it would be possible to manage DVFS and power-gating directly from inside my application.
> >>
> >> Sorry for all the questions, but any help would be very much appreciated.
> >>
> >> Thanks in advance for your kind attention,
> >>
> >> Best regards,
> >>
> >> Francesco Comaschi
> > _______________________________________________
> > linaro-kernel mailing list
> > linaro-kernel@lists.linaro.org
> > http://lists.linaro.org/mailman/listinfo/linaro-kernel
> >

power measurement

[Grover, Andrew]

> [mailto:cpufreq-bounces@www.linux.org.uk] On Behalf Of Todd Poynor
> Amps consumed should be a good predictor of expected battery life, 
> independent of the vagaries of any particular battery.  The expensive 
> scopemeter used for the attached measurements isn't really 
> necessary, by 
> the way; a $130 multimeter that averages DC amps, placed in series 
> between the DC transformer and the notebook, should do fine.  In case 
> this helps,

Can we all get into the habit of reporting power numbers in Watts? Just
take the amps and multiply by the voltage. This lets one better compare
platforms with differing supply voltages.

Please also make sure you have enabled ACPI processor C states. This
will help to lower idle CPU power in whatever performance state you are
in.

Regards -- Andy

Re: power measurement

[Russell King]

On Thu, Oct 23, 2003 at 12:48:23PM -0700, Grover, Andrew wrote:
> > [mailto:cpufreq-bounces@www.linux.org.uk] On Behalf Of Todd Poynor
> > Amps consumed should be a good predictor of expected battery life, 
> > independent of the vagaries of any particular battery.  The expensive 
> > scopemeter used for the attached measurements isn't really 
> > necessary, by 
> > the way; a $130 multimeter that averages DC amps, placed in series 
> > between the DC transformer and the notebook, should do fine.  In case 
> > this helps,
> 
> Can we all get into the habit of reporting power numbers in Watts? Just
> take the amps and multiply by the voltage. This lets one better compare
> platforms with differing supply voltages.
> 
> Please also make sure you have enabled ACPI processor C states. This
> will help to lower idle CPU power in whatever performance state you are
> in.

Also note that the battery charger state should be defined when measuring
the power supplied to a laptop from its AC adapter.  If one laptop is
charging the battery when you make a measurement and another isn't,
you won't be comparing like with like.

I'd also suggest that there will be a fair variability between the
battery chargers in laptops, so this type of measurement will only
make sense if the charger isn't running.

(Note that certainly IBM thinkpads switch to charge mode for a short
time, even if the battery is 100% when the AC adapter is connected.)

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 PCMCIA      - http://pcmcia.arm.linux.org.uk/
                 2.6 Serial core