A Plethora Of Linux Power Tests Are On The Way

[AlbertP]

At datacenters, they don't use much laptops. A desktop or server does not have the advanced power management a laptop has, and is affected much less.

[chithanh]

It is even worse. Imagine the following hypothetical and completely made up situation: The system firmware can perform function X using interrupts or busy-waiting. Using busy-waiting, performance in GPU-bound 3dMark06 is 0.1% higher, but power consumption increases due to CPU load. In order to look better in benchmarks, the manufacturer switches between interrupts while on battery power and busy-waiting while on AC power.
Now two regressions are independently introduced in the kernel, regression A causes more power consumption for the interrupt case and regression B causes more power consumption in the busy-wait case.

What mobile users care about is regression A, but testing on AC power only finds regression B.

When you are running 400+ linux servers in a datacenter, you don't want to upgrade to something that will multiply your power usage draw.

for a laptop it goes up a couple watts, for a datacenter it goes up a kilowatt or more...

For datacenters, you need to plan power supply and cooling according to TDP/peak power consumption anyway, so I don't think it will have a dramatic impact.

[Lederhosen]

You correctly state that "major kernel power regressions that are hitting many mobile Linux users", however you proceed to measure the power draw on desktop systems. Whether the results can be transfered to mobile setups is questionable and would have to be verified there anyway.

I think the most meaningful results you get from directly measuring the voltage and current coming from a notebook battery. That way you avoid both the inefficiency in the power supply and the inefficiency in the charging logic, both of which don't really matter to mobile users but still skew the results.

for the purposes of the power usage benchmarking, I don't see a need for accurate measurements of the laptop sans the power supply.

power draw will still increase from the wall when using the newer kernels that aren't regulating the system properly.

for that matter, this has a lot of interest for servers as well as mobile. When you are running 400+ linux servers in a datacenter, you don't want to upgrade to something that will multiply your power usage draw.

for a laptop it goes up a couple watts, for a datacenter it goes up a kilowatt or more...

[Lederhosen]

You have to be aware that there are tools in Linux that will adjust laptop hardware to run at different energy saving level when running plugged in vs on battery so they may give you vastly different wattage results

laptop-mode-tools is one example.

All very true, but you can force on that functionality to enable the features. Laptop mode tools allows you to even set it to power save mode full time in the config file.

[AlbertP]

And many hardware has a power saving mode used when running on battery power, e.g. wireless network cards.
I can just see some network quite far away from me when running on AC power, but with battery power I only see networks from the same street.

[FunkyRider]

You have to be aware that there are tools in Linux that will adjust laptop hardware to run at different energy saving level when running plugged in vs on battery so they may give you vastly different wattage results

laptop-mode-tools is one example.

[chithanh]

Measuring on the DC side of the power supply is definitely better than measuring on the AC side, but still you have to keep in mind that there is no incentive for the manufacturer to be particularly efficient there. Plus, whether the power draw from DC side is proportional to battery power draw seems somewhat plausible, but would have to be verified on a case-by-case basis.

[sturmflut]

I think the most meaningful results you get from directly measuring the voltage and current coming from a notebook battery. That way you avoid both the inefficiency in the power supply and the inefficiency in the charging logic, both of which don't really matter to mobile users but still skew the results.

..or just unplug the battery and use one or two USB multimeters to measure voltage and amperage between the power adaptor and the laptop, otherwise you have to continually recharge the battery and can't take any measurements during that time . The measurement is way more accurate too, and an USB multimeter comes for less than 45 $.

[chithanh]

You correctly state that "major kernel power regressions that are hitting many mobile Linux users", however you proceed to measure the power draw on desktop systems. Whether the results can be transfered to mobile setups is questionable and would have to be verified there anyway.

I think the most meaningful results you get from directly measuring the voltage and current coming from a notebook battery. That way you avoid both the inefficiency in the power supply and the inefficiency in the charging logic, both of which don't really matter to mobile users but still skew the results.

[curaga]

Heh, eagerly awaiting the LKML mail listing 10+ commits causing power regressions

Since apparently nobody else is measuring this, I don't doubt that you'll find many more than just the mentioned two.