Mobile Users Beware: Linux Has Major Power Regression

[plgaming]

@Michael:

Is it possible to use PTS to do an automated git bisect? The Timed Idle and OpenSSL have very few fluctuation, so maybe

(median currentCommit) / (median 2.6.37) > 1.05 => git bisect bad

If we are lucky, the problem is a single commit and we can sort out the issue.

[Michael]

@Michael:

Is it possible to use PTS to do an automated git bisect? The Timed Idle and OpenSSL have very few fluctuation, so maybe

(median currentCommit) / (median 2.6.37) > 1.05 => git bisect bad

If we are lucky, the problem is a single commit and we can sort out the issue.

Right, PTS can handle it but on the last page of the article I mention the current problems in doing so.

[deanjo]

It seems you have no clue of what the real world is about. All laptop manufacturer only care about making their product work on windows and for that they do tons of quirks in the driver instead of fixing their bios & acpi, so their driver endup with a tons of special tweak specific to one model and one model only.

Apple does the same, they control the hardware and the software but most of the time they do a tons of quirk inside the operating system to improve how it works, i have seen some of the osx kernel and it's populated with special path for each specific hardware, it's a mess.

Linux kernel on the other end try to do cleanly in a generic way, and on that front i am sure linux is a good contender. Of course in real world people have crap hardware with crap acpi and crappier bios, but there is just not enough resource to go after all the hardware out there and fine tune each of them.

A "mess" that works > "neat but non-functional". That my friend is what "the real world" is.

[crumja]

This is a real issue. I doubt that some silly PTS config or faulty default resulted in consistently higher power usage across the board on a variety of testing platforms. A true bisect of the kernel should be done to find the commit at fault. The problem is that no one cares on the dev team!

I have a suspicion that this event highlights the dearth of varied testing that the kernel is subject to. Developers are users with pretty homogeneous usage patterns. Remember Ted Tso's incident with ext4 a few years back? Ext4 was deleting data like crazy on reboots and it turns out no one in the development community bothered to test that case because none of them *ever* had to do a hard reset for any reason.

The problem with responsiveness under HD activity that Con Kolivas brought up a few years ago was never fixed until 2.6.38. No one would even admit to having that problem because they all ran 16+ core machines with multiple hard drives and SSDs that RAID away any perceptible performance problem. I remember accusatory "show me the data" statements from devs when users around the world all confirmed the problem. Kernel development is more geared towards enterprise applications today because that's who pays developers and that's where money is made from linux.

Historically, power consumption has always been a problem with the linux kernel. Developers either aren't in tune with user demands (probably because they sit at desktops all day long or buy extra extended batteries to use with their Thinkpads, which are the only notebook any one of them would even consider) or just don't think of it as high priority (again, reflecting the enterprise slant of development efforts).

[plgaming]

Right, PTS can handle it but on the last page of the article I mention the current problems in doing so.

https://twitter.com/#!/michaellarabe...59835757522945

That was fast! I hope will be found with ease.

[smitty3268]

A "mess" that works > "neat but non-functional". That my friend is what "the real world" is.

Call me back when OSX works on the hardware I own. Until then, "neat but non-functional" is a lot better than "won't work at all". Which is crap, actually, given how many servers out there run linux.

[codedivine]

I think power consumption is not obtained from hardware but calculated by ACPI. It's rather simple mathematics, but it could be the algorithm has been changed in between.
If you could try a full battery emptying between, let's say 2.6.37 and 2.6.38 we could be sure about the reported consumption.

+1 to this.

[airlied]

This is a real issue. I doubt that some silly PTS config or faulty default resulted in consistently higher power usage across the board on a variety of testing platforms. A true bisect of the kernel should be done to find the commit at fault. The problem is that no one cares on the dev team!

I have a suspicion that this event highlights the dearth of varied testing that the kernel is subject to. Developers are users with pretty homogeneous usage patterns. Remember Ted Tso's incident with ext4 a few years back? Ext4 was deleting data like crazy on reboots and it turns out no one in the development community bothered to test that case because none of them *ever* had to do a hard reset for any reason.

The problem with responsiveness under HD activity that Con Kolivas brought up a few years ago was never fixed until 2.6.38. No one would even admit to having that problem because they all ran 16+ core machines with multiple hard drives and SSDs that RAID away any perceptible performance problem. I remember accusatory "show me the data" statements from devs when users around the world all confirmed the problem. Kernel development is more geared towards enterprise applications today because that's who pays developers and that's where money is made from linux.

Historically, power consumption has always been a problem with the linux kernel. Developers either aren't in tune with user demands (probably because they sit at desktops all day long or buy extra extended batteries to use with their Thinkpads, which are the only notebook any one of them would even consider) or just don't think of it as high priority (again, reflecting the enterprise slant of development efforts).

Thats nearly as good a generalisation as the one about only clueless idiots post on forums...

Dave.

[kraftman]

It boggles the mind how after so many years the rest of the world is still trying to catch up to OS X's power management. You would think in this day and age that it would be common knowledge how to do it properly and not be some black obelisk.

At first we need to know if this test was done right. Btw. it doesn't seem Android has some power management problems, so I wonder if os x does pm better.

[kraftman]

Thats nearly as good a generalisation as the one about only clueless idiots post on forums...

Dave.

It's not even worth to quoting such idiocy.