Linux Kernel Suffering Power Management Regression? 186
Posted
by
Soulskill
from the year-of-linux-on-the-coppertop dept.
from the year-of-linux-on-the-coppertop dept.
An anonymous reader writes "It appears that there's a big power management regression in the Linux kernel for the 2.6.38 and 2.6.39 development releases, including the kernel to ship with Ubuntu 11.04 next week. It's reportedly causing a 10~30% increase in power consumption on many laptop computers."
Only Power Users will notice (Score:5, Funny)
Re: (Score:1)
I think this is something that only Power Users will notice. It's not something important for the common user.
I think a lot of people would notice if their laptop suddenly got a third less battery life.
Re:Only Power Users will notice (Score:5, Insightful)
I think this is something that only Power Users will notice. It's not something important for the common user.
I think a lot of people would notice if their laptop suddenly got a third less battery life.
Excuse me sir, I believe your pun detector is broken.
Re:Only Power Users will notice (Score:5, Funny)
He might not have the capacity to appreciate such humor.
Re: (Score:2)
He might not have the capacity to appreciate such humor.
No, the GP is right. The battery in my pun detector was dead.
Re: (Score:3, Funny)
your continued resistance to puns is growing alarming
Re: (Score:2)
Re: (Score:2)
I might suggest your not clipping them to your nipples.
Re: (Score:2)
I dunno, I'm getting a bit of a charge out of it.
Re: (Score:2)
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
Geek alert! Geek alert! Geek alert!
Congratulations, you have detected a geek on a website that labels itself "News for nerds [xkcd.com]".
Re: (Score:2)
I don't see how masturbation would help. Ever try programming when some of your keys no longer work? It makes for very interesting variable names.
But I guess real programmers lay a sheet of saran wrap over their keyboard before they start fapping.
Re: (Score:2)
http://xkcd.com/378/ [xkcd.com]
No worries for real programmers...
Re: (Score:1)
Which insensitive sod has modded parent down?
Re: (Score:1)
Actually I would guess it was an overly sensitive sod.
Re: (Score:2, Offtopic)
Which insensitive sod has modded parent down?
That's insensitive CLOD , you insensitive twit.
Re: (Score:2)
Re:Only Power Users will notice (Score:4, Informative)
Linux on laptop (Score:3, Informative)
Why would anyone want to run linux on a laptop? Well, I run linux on my laptop. At first I had it set up to dual boot, but after months of not using the Windows partition I canned it and have never missed it. During the period I had both operating systems set up, I could compare them. Windows (Vista as installed at the factory) was dog slow and buggy (and before you poke fun at Vista, XP was just as bad on other laptops as received). Linux was snappy, remarkably stable, and supported the hardware very
Re: (Score:2, Interesting)
Oh I'm not saying you can't do it, it just seems like a hell of a lot of work for very little gain. Personally I wouldn't wish Windows Vista on my worst enemy (although win 7 is quite nice, as is XP without the crapware) but the problem is unless you are an uber geek laptops are a mess and frankly the odds they'll even get 40% of them running stable is a big question mark.
Plus this is probably THE absolute WORST time in history to be working on the things, as we are talking a HUGE fundamental shift in the w
Re: (Score:2)
Yeah, Windows XP. That's why half the time I install it on a machine made after 2002, it doesn't have (among other things) the NIC driver, which means no Windows Update. Solution? Boot Ubuntu LiveCD and download it from the internet -- works every time.
Re: (Score:3)
That is a most interesting link. I pretty much agree with Linus on everything. You want formal design, you get hurd. Or rather you DON'T get hurd. Not in our lifetime (mine anyways). You follow Linus' evolving ideas and methodology and you get an excellent working product which evolves better and better.
All I can say about the guy in the basement is, yeah, there are lots of them, and yes, they keep up fine with Linus' ways and the difficulties you mention. I can't explain to you how this all works, be
Re: (Score:2)
Oh I'm not saying you can't do it, it just seems like a hell of a lot of work for very little gain.
Like... putting the CD in the drive, booting up and selecting 'install'?
I know there are problems with recent Nvidia chips which require undocumented hacks due to their video switching between the integrated crap and the Nvidia GPU, but when I installed Ubuntu on my Toshiba last year it all just worked.
Re: (Score:2)
Oh and FYI you don't actually have to track down drivers anymore, Windows Update takes care of that for you.
Oh, does it? Strange. The last time I installed Windows 7 I had to fidge around about an hour just to get it to recognise the SATA controller. And I still have to hunt down driver updates for my mouse, my video card and my chipset myself. And the mouse is the only item even remotely esoteric. On Linux, on the other hand, the only thing I have to take care of myself is the NVIDIA graphics driver, and that only because I run beta drivers and therefore do not want to use the built-in update mechanism.
Re: (Score:2)
He's trollin'. If you're using a reasonably usability-oriented distro (e.g. Fedora, Ubuntu, Mint), you'll have virtually no issues.
True. Debian runs on mine like a champ from the get-go, wireless and all. The only thing that troubles me is Intel's video driver. They just can't or won't write decent drivers for Linux, so performance takes a hit.
Re: (Score:2)
Debian has been running so well on my 6 year old Thinkpad X41 that I haven't even considered buying a new laptop. I'll probably use it as long as the hardware lasts, be it another 4 or 6 years. After upgrading to SSD and aligning the partitions correctly to erase byte size it boots up to desktop (Fluxbox) in 7 seconds and basically everything that doesn't require heavy calculation from the CPU happens instantly. In practice it's much faster in all CPU non-heavy actions than any Windows laptop I've ever used
Re: (Score:2)
Ditto. I've been running Ubuntu on my Thinkpad T43 for 5 years already, currently Ubuntu 9.10, ditching Windows and never having to look back. Thinkpads and Linux are almost made for eachother :) This is to the person who was complaining how laptop hardware is so exotic that Linux doesn't know what to do with it. Thinkpads (mostly) ship with reference Intel Centrino implementation, or whatever Intel calls their "notebook platform" these days, and as far as Linux compatibility goes, you HARDLY can do better.
Re: (Score:2)
Intel drivers used to work great just few years back. IIRC it was a major rewrite of Xorg and introducing KMS that has caused the regression and the need for rewriting much of the driver code. (anyone who knows better please correct me). I think worst of that is already behind and the drivers are improving even for my old i915 chip. Some font or other rendering artifac
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Intel is a commercial enterprise. They don't need to write drivers if they have no incentive to do so. If they write drivers for something like Linux, it's simply because they have a hand in it.
What they arguably DO NEED to do is provide a satisfactory specification of their hardware, so that the people who shout "open source! everyone can contribute" can put theirselves where their mouth is and start well, producing decent drivers for Intel hardware. As simple as that. And Intel did provide EXCELLENT docum
Re: (Score:2)
So your answers are, in no particular order, YouDisagreeWithMeThereforeYou AreATroll [tmrepository.com] along with WorksForMeAtHomeItDoesReally [tmrepository.com] and ImaginaryProblemsKillWindows [tmrepository.com] with a bit of LinuxSupportsMoreHardwareThan Windows [tmrepository.com] when in reality you get ThisIsAKnownProblemNowGetLost [tmrepository.com] and YouAreUsingTheWrongHardware [tmrepository.com]
Isn't it kinda sad that I can answer EVERY SINGLE THING you and the other FOSS zealots here say with TMs? Doesn't it bother you, even just a little? You know why I can do that so easily? It is because you have been
Re: (Score:3)
After installing Ubuntu every half-a-year for the fourth time, I was getting tired of "bridging" the gap between an out-of-the-box Ubuntu state to the one I prefer, so I simply wrote a bash script that does everything :) I mean it sets all my GNOME settings (through gconf- command line tools), sets up launchers and mime type preferences, icons, installs applications, even downloads and compiles stuff that cannot be found in repositories or in case I prefer a compiled version. Basically, when the script is r
Re: (Score:2)
Maybe 5-10 years ago there were problems running Linux on some laptops but now a days it generally just works. I've run openSUSE and SUSE Linux Enterprise Desktop on several Dell (Inspiron and Latitude), one IBM (Lenova) laptop and one Toshiba laptop with minimal or no trouble. A very long time ago (8+ years) I had a Dell Inspiron laptop which had issues with its internal wireless card but I just popped in a PCMCIA wireless card with Linux support until the internal wireless card was supported.
Re: (Score:2)
Re:Only Power Users will notice (Score:4, Informative)
vista came on this laptop. there are NO xp drivers.
There are actually entire lines of computers that came with Vista but were too slow to actually run either it or Windows 7 properly, and at the same time are too new for anyone to have made XP drivers. Like half the computers sold with Vista before 2009 or so.
Never had any problems running Ubuntu on them though.
Re: (Score:2)
Nothing that runs Vista will run Windows 7 worse. Windows 7 is better in low memory situations. So you have just spoken complete and utter nonsense. Only your Ubuntu line at the end stops me from suspecting you of being a Microsoft shill. I have one of the machines you're talking about, it's a Gateway netbook with an Athlon 64 L110 CPU and R690M chipset, very sweetly low power use under Vista. Unfortunately AMD did not contribute power saving to Linux so I get about half the battery life there. Under Window
Re: (Score:2)
I am going to quote what I wrote so that you can read it again:
There are actually entire lines of computers that came with Vista but were too slow to actually run either it or Windows 7 properly
Windows 7 may be faster on low memory machines than Vista, but that's like saying that a turtle is faster than a slab of concrete. The problem is that there are zero versions of Windows that run properly on low resource machines with no XP drivers.
Even the ones with XP drivers are often hopeless. I've seen XP run on a Pentium M from that era -- which it does admirably for exactly as long as you don't install antivirus on it, after which point th
Re: (Score:2)
Well, I do feel a little silly having read your comment, although I've also read a lot of reports of people running Windows 7 successfully on pretty limited hardware. Long boot and application load times, but otherwise no big deal.
I do agree that lots of machines have shipped with less RAM that is really necessary. I've done the 1GB->2GB shuffle on two netbooks and a desktop so far. The desktop came with Vista. It actually claims to max out at 2GB, it's a Gateway GT5475E and I forget who made the mainboa
Re: (Score:2)
Anyway, your point about the machines is taken, but all they need is more RAM.
Sure, assuming you're the Slashdot reader willing to install it yourself rather than Joe Sixpack who has to pay Best Buy $200 to add $40 worth of memory to a $150 computer. And assuming it's even upgradeable -- some of the cheaper laptops actually have the memory chips soldered to the motherboard. Or they just don't support anything more: A 1.4GHz Pentium III-S will still thoroughly embarrass almost anything Atom-based, but a lot of those motherboards max out at either 512MB or 1GB.
Re: (Score:1)
Re: (Score:2)
Quite the reverse in fact. Power users care about how much CPU power they can use. Common users care that they can get to $city on the train without their laptop running out of battery.
Re: (Score:2)
power or common really doesn't matter, as a common user of a craptop that has to be plugged to the wall anyway, I wouldn't notice.
Linus Torvalds and regression? (Score:1)
Re: (Score:3)
Re:Linus Torvalds and regression? (Score:5, Interesting)
Ever notice it is only Phoronix reporting that?
When did steam come to linux again?
Sorry, but I want to see this backed up another source before I just go believing it.
Re:Linus Torvalds and regression? (Score:5, Interesting)
Ever notice it is only Phoronix reporting that?
Do you know of any other organization with a large automated regression testing system for linux kernels? That's not just me being snarky, its a serious question - who else beside phoronix is doing this sort of wide-scale testing on a constant basis?
Re: (Score:2)
Do you know of any other organization with a large automated regression testing system for linux kernels? That's not just me being snarky, its a serious question - who else beside phoronix is doing this sort of wide-scale testing on a constant basis?
I can't answer your question, but Phoronix's testing and benchmarking is notoriously bad, and anything you see there needs to be taken with a gigantic nugget of salt. At best, it's a hint to look around and see if you can find any similar result from a more reputable source.
Re: (Score:2)
What do you base your criticism of Phoronix on, exactly? At best you draw attention to yourself and not to how bad they are or are not. I've been reading them for a couple of years and even though perhaps they haven't won awards for their journalism or website usability, the information provided SEEMS credible enough and worth a read. Certainly not "notoriously bad" or to be taken with a "gigantic nugget of salt".
Maybe YOU could build and benchmark a couple of stock kernels on a simple-as-brick Thinkpad T60
Re: (Score:2)
What do you base your criticism of Phoronix on, exactly?
How about because they're the Daily Mail of tech news? Did you sleep through the years of bullshit about Steam and UT3? Ever noticed the fact they sound the klaxons five minutes after any minor bug in a 2.6.X-rcY kernel is found? How about the fifty pages of graphs per article just to deliver fifty times as many ads?
Because that's what it all boils down to: Phoronix only exists for the sole purpose of trolling sites like Slashdot so they can rake in the cash. If I wanted hardware benchmarks there's no short
Re: (Score:2)
They do. But then the reports get through to the managers, and they feed the papers to their closest office paper shredder or archive the emails, and proceed to write cryptic blog entries on how they had to scrap feature X because users wouldn't care or something like that.
Re: (Score:2)
Ever notice it is only Phoronix reporting that?
When did steam come to linux again?
Wine 0.9.6x, as I recall...
Re: (Score:2)
Now redo that just updating the kernel and get back to us.
Re: (Score:2)
Do you run the Unity thing? It uses Compiz, which uses some multimedia timers and the graphic hardware, as opposed to plain good old Metacity on Ubuntu distributions prior to 11.04 which draws stuff using simpler hardware subsystems and no fancy timers etc. Just 5W seems even too good, i'd imagine Unity+Compiz drawing a bit more of a difference.
Re: (Score:2)
I run my recently upgraded desktop (Core i5-2500K) through a power meter and it is very easy to notice - Ubuntu 10.10 (kernel 2.6.35) idles at about 30W, while 11.04 (2.6.38) goes up to 35W. That's about 20% more.
I believe 2.6.36 introduced a problem with DRI on intel and some other video chipsets which led to a shitload of kworker wake ups. One of my laptops running Arch benefitted from a kernel parameters suggestion over here [archlinux.org] but by that stage I had a menu.lst as long as your arm full of attempts at a functional system (e.g. kernel ... nohz=off highres=off pcie_ports=compat...) each with small, incremental improvements.
You can't catch every regression without owning all hardware and infinite time :)
Re: (Score:2)
That's about 20% more.
No. (35-30)/30 = 5/30 = 1/6 =~ 16.7%.
Or roughly; 20%
Re:Linus Torvalds and regression? (Score:5, Informative)
Well, what Linus is focused on is breaking user code - if it worked in a released kernel, you will not break it in any future kernel. I don't think there's any strict rule that performance must always be better or power consumption lower. Particularly if you're not doing something "right" and have to add additional checks/locks/synchronization for corner cases that can slow you down, they generally value correctness over performance. That's the case in many of Phoronix' sensationalist news, a development release is very fast but when you make it work 'right' the performance is no longer that impressive. That stuff will happen as close to the bleeding edge as most of the things they report on are. Of course, they do find real regressions too but it's easy to get the wrong impression...
Re: (Score:2, Interesting)
I'm on .38 on multiple computers and I'm not seeing these issues, perhaps it's a configuration error on their end (assuming they compile the kernel themsleves for testing) or a configuration error on whatever distribution they test with?
Re: (Score:2)
Yes, a 38 police special will do a dandy job on any computer!
Not true. I remember about ten years ago someone talking about a demo they saw of a fault-tolerant computer where the salesman emptied a .45 automatic into the computer while it was running and it continued working happily despite the damage.
Re: (Score:2)
Re:Linus Torvalds and regression? (Score:5, Interesting)
Re: (Score:3)
One would expect that there is some sort of automated performance regression suite that is run regularly (say, daily) to catch offending commits.
Re: (Score:2)
Not just the CPU (Score:2)
My radeon card no longer has functioning power management, as of 2.6.37. It used to be possible to echo dynpm to a sysfs interface and it would downclock the card. That no longer does anything. I did send in the bug to the maintainer, but it apparently is not a high priority item.
Re: (Score:2)
Re: (Score:2)
Perhaps it is that Wayland nonsense eating up more CPU cycles and using less of the GPU.
Re: (Score:2)
Re: (Score:2)
Pah! Why on earth would a kernel be doing more work? That means they're doing something more inefficiently, that's a regression.
I didn't say it wasn't a regression. I said it probably wasn't an intentional regression (i.e. a change made knowing it would cause incompatibility or inefficiency with things that were compatible/efficient before).
It may not even be that more work is being done at all: it could be a problem somewhere in hardware initialisation code that is causing something to not have its power saving mode activated where it previously did: it is not at all inconceivable for a bit of code that usually runs once at boot
Re: (Score:3)
run lm-profiler (part of laptop mode tools if I recall correctly) and see what demands that your hard drive be awoken from sleep :)
Re: (Score:2)
Remember that when 2.6 came out, Torvalds himself decided to stop keeping separate stable and development kernel trees. Arguably, there wasn't a lot of point - most Linux distributions add all sorts of patches anyhow, it wouldn't kill them to add another patch here and there if such a regression does come up.
So now, such a regression does come up occasionally. Doesn't matter if it does - it's not like RedHat or SuSE are going to ship the latest 2.6 kernel without testing and patching where appropriate any
Depends on the type of regression (Score:2)
The reality of the "No regressions" phrase a bit more complicated. It help tremendously if the regression is noticed during the development stage of the kernel and can be narrowed down to a single patch. Further it also depends on exactly which way you regress - a "can't boot previously working and still supported system" functional regression can be treated differently to a "inserting a new disk is 0.1 of a second slower" performance regression.
Re: (Score:1)
Yes, that's true. But in all fairness it has happened to other systems as well. Bill Gates didn't like regression and some guy left because of an argument concerning some type of regression.
Stuff happens.
The real test is seeing what people do AFTER "stuff happens".
moronix. again. (Score:1)
nothing to see here, move along
Moronix test suite (Score:3, Interesting)
Overheating probs (Score:3)
Well, that would explain a great deal why my Dell laptop has been overheating and shutting off without warning since that last kernel build. It's shut off three times today and I haven't even done any intensive computing.
Methinks I need a new box.
Re:Overheating probs (Score:5, Insightful)
>you think you need a new machine
Either that or open it up and blow out the dust.
Works wonders for overheating, dontchaknow.
Cheap/easy fixes first. Always.
--
BMO
Re: (Score:3)
Exactly. I had a Latitude D830 whose CPU was running at a more or less constant 212-218F and was virtually unresponsive. After blasting some canned air into the vent on the sides and back it started working like new.
I talked to the help desk guys about it and it's a pretty common occurrence with Dell laptops. Seems like a major design flaw to me.
Re: (Score:2)
> CPU was running at a more or less constant 212-218F
I have machines with fans either dead or completely removed that don't run that hot.
Re: (Score:2)
It's what the system monitor reported. I usually run VirtualBox with Windows, NetBeans, JBoss, and run maven builds constantly, so it was pretty busy. I burned my finger when I touched the vent to feel how hot the air was.
Re: (Score:2)
Get yourself an IR thermometer. You don't need to trust what the software is telling you.
Re: (Score:2)
Um, 212F is 100C last I checked...
Re: (Score:2)
My three thousand dollar HP laptop choked itself with dust, too. It's not a problem with Dell laptops. It is a major design flaw. It's not clear what can be done about it.
Re: (Score:2)
Granted, but I had the exact same problem crop up yesterday *immediately after upgrading the kernel*. It could be coincidence but if enough people suddenly discover they have to clean out their fan heatsink then there's probably an additional factor at work.
Re: (Score:2)
You might need to replace the fans. I need to do that every 2 years or so in my latest ThinkPad. It is the only thing that breaks, but it is starting to fail for the second time now, just as the laptop is turning 4. The laptop is still fast enough, so to replace fan or replace laptop?. Fortunately the fan only cost $100 to replace with official spare parts and Lenovo has a nice official online guide on how to do it
Re: (Score:2)
It is laptop fan for the cpu and graphics processor. It is one piece of custom shaped aluminium and copper with a small fan embedded inside the cast. I actually think it is a fair price for a custom replacement part. Though I could check again if the fan inside the piece is reachable, but since the issue is that it gets looser and not that it breaks, I doubt I can make it better.
Well... here's a confirmed recent bug. (Score:3)
Power consumption raised significantly in natty [launchpad.net]
this is the actual confirmed (4-13) bug report on the Launchpad at least a particular instance.
Personally I do not run the extra baggage of Ubuntu on my mobile linux device. (netbook)
When did they start putting unconfirmed or untested bug reports on Slashdot? Sure TFA says much to warrant further investigation... but not to have people like me get curious. (Just my opinion)
Re: (Score:2)
I may be referring to the wrong distro, oops.
Re: (Score:2)
It actually is the right distro tyvm.
And yes I am inexperienced in these matters...and I do it for the ads in public terminals.
Re: (Score:2)
Personally I do not run the extra baggage of Ubuntu on my mobile linux device. (netbook)
Personally I run ubuntu-minimal on my EEE 701. Best experience I've had with a computer, period, since Jolicloud fucked up their launcher. Everything works, and works flawlessly all the time. I only wish I had a faster machine which worked so well.
Assumptions... (Score:2)
At least run powertop and compare the output of lspci -vv for each kernel.
FreeBSD (Score:2)
A little off-topic, perhaps, but how are some of the other *nix's doing in this respect, such as FreeBSD? Is FreeBSD even valid option for a laptop or a netbook?
CAN WE STOP POSTING THIS VILE PHORONIX CRAP!? (Score:4, Interesting)
Phoronix is shit. Pure, grade-A shit. Worthless.
They have _nothing_ of value to add to anything. Sensationalist crap which is not reported elsewhere, _because it it not an issue_.
Regressions in the development kernels are part of the process. Even actively trying to avoid Phoronix, I have seen tons of those non-news about some random regression and the breathless follow-up that, lo surprise, they didn't just release but fix the issue. Woooooo!
Phoronix is shit and it should be blacklisted globally on Slashdot and anywhere else. Stop linking to them, stop commenting on them (other than making others aware of this).
Rant over.
Re: (Score:2)
Slashdot article says development kernel, not stable. So does the parent.
how about basic features (Score:5, Insightful)
Re: (Score:2)
With MS's track record I think they're their own prior art.
Re: (Score:2)
No, he is Anonymous Coward. It is not clear if he reads anything (other the two words that he quoted, of course).
Re: (Score:3)
Yes, it's probably due to that one thing you heard about.
Re: (Score:2)
Measuring heat may not be a fair comparison. Perhaps the 2.6.38 performs >20% better at that higher temp? How many loops are you getting in that period?
I have a general perception that running fast for a short time and idling at low power for longer is generally better for low power consumption.
Also heat may not necessarily be a good indication of performance directly. Just as battery life is not necessarily a direct indication of performance.
Correlation != causation.
ws