Slashdot Mirror
Intel's PowerTOP Extends Linux Battery Life
DuracellFan writes "Intel recently released its PowerTOP utility, which builds on work done by kernel developers to make the Linux kernel power-efficient. PowerTOP gives a snapshot of what apps are consuming the most power. The PowerTOP website also hosts patches for several Linux apps and the kernel. In the Linux.com article, lead PowerTOP developer Arjan van de Ven of Intel says that PowerTOP could soon show which applications keep the disk busy." Linux.com and Slashdot are both part of OSTG.
The hardware that runs it does! Typical Intel, trying to shift the blame.
:)
Of course this utility is very useful for developers and for Linux distributors.
For the average user it is a nightmare.
The blurb says that the tool told him to disable beagled which he did and he was duely impressed when the number of wake ups per second dropped. However the actual watts used went up. Thought the point was to save power?
No wonder people say Linux has bad driver support. This is like running windows98 today and claiming that modern devices have no drivers.
Even RHEL and Debian stable, which make up a huge chunk of enterprise server linux in the USA use 2.6 kernels.
Amazing - if I have an application which chews CPU time, does I/O, or a peripheral which powers another or requires a high power output like a wireless card, then these applications or peripherals drive power consumption up!
Brilliant!
I want to delete my account but Slashdot doesn't allow it.
I have a server at home that I'm about to upgrade the kernel on, and I would very much like to have it as energy efficient as possible... but I worry about stability. Is NO_HZ safe?
Belief is the currency of delusion.
Read up a bit, it's about applications behaving in energy inefficient ways (waking the CPU) when they could get the same work done without waking the CPU as often. This concerns polling loops and such, as well as things you normally don't think about, like animated cursors. powertop helps you track down which applications are behaving badly, so that they can be fixed.
Belief is the currency of delusion.
Slackware has even shifted to 2.6.x
No one here gets out alive
DEAR User,
Sorry but you're mistaken.
You actually discovered our latest feature.
You haven't read about it yet, because we were developing and testing it until very recently, and we didn't want to speak to early about it.
We, as developers conscious of their travelling users, that have so much time that they need to work as they are in the train, have though of YOU !
As such we present you our latest feature :
WE GIVE YOU THE POSSIBILITY TO COOK YOUR DINNER ON YOUR LAPTOP (so you can do even more important things during the time you're commuting, which will leave you more free time when you reach your destination !)
Alternatively, you can also use our application on your laptop as AN INCREDIBLE AND COMPACT LAP-WARMER !!! For all those long commute during winter.
(DISCLAIMER : Warning, do not use with Batteries manufactured by Sony).
Thank you, wish you enjoy our brand new features.
- The Dev team.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
C'mon what are we talking about here, a few minutes? AFAIK, better power savings comes through a good acpi config, which I don't see a whole lot of discussion on.
My guess is where this kind of thing would make a dollars/cents difference is in the NOC. But this kind of detail isn't very sexy or very high on most NOC operators radar.
http://www.maxineudall.com/2010/02/should-economists-be-sued-for-malpractice.html
"We are proud to announce that Slackware Linux version 11.0 has been finally released; it took some time but is well worth the wait. You can read the official announcement at this link. Slackware 11.0 contains the 2.4.33.3 Linux kernel, 2.6.17.13 in /extra and 2.6.18 in the /testing directory; the default boot option is the dependable 2.4.33.3 kernel, but this time we included more choices and support for 2.6.x kernels..."
http://www.slackware.com/
It could mean as much as an hour or two, depending. The less the CPU sleeps, the more power it consumes. The more the HDD gets accessed, the more power it consumes. ACPI doesn't buy you much if your CPU is constantly running at full clock and your HDD is always spinning.
My blog
Success Stories
Guess you could accuse him of bias...
Belief is the currency of delusion.
That depends. Laptops are saving power because presumably they're idle most of the time and this program can tell you which processes behave badly while "idle" (by, say, polling the HDD for no good reason). On a server presumably your machine spends very little time idle (since you're serving stuff), so there isn't much opportunity for power savings from an application like this.
I read the internet for the articles.
BUT - on a desktop, or laptop? Nuh-uh. I'm kinda greedy about functionality and performance in those cases.
Quo usque tandem abutere, Nimbus, patientia nostra?
I'm a little skeptical about that value (but willing to be convinced). I only pay about $300/year for electricity overall. Of course, I don't own a PS3. However, I do have a refrigerator and an air conditioner (as well as other devices that use electricity).
Ben Hocking
Need a professional organizer?
Most people i know still run 2.4.x and Slackware still ships with 2.4 as default.
Slackware users don't know people. Stop lying! It's just yourself who runs 2.4, right?
AMANDA, fixes that issue.
It is free and Free. My boss liked the free part.
Wow. That's really cool. I love how open source allows you to tweak your settings down to the core like this - and Intel is the company that made it.
in girum imus nocte et consumimur igni
Power consumption IMHO is not a problem peculiar to laptops. And im sure a lot of machines exist out there running kernels 2.6.2 that wont be able to use this utility. Considering that slackware isnt primarily targetted at servers or business machines, i would say the fact that it ships with the 2.4 kernel as default means quite a numbe rof hobbyists, students and home users still run 2.4.
C'mon what are we talking about here, a few minutes? AFAIK, better power savings comes through a good acpi config, which I don't see a whole lot of discussion on.
Do you even know what ACPI is? Have you read the link? (clearly not)
No matter how well your "acpi config" is done, if you've a process eating 100% of the cpu power all the time, your batteries will last less than a compuer with no ACPI that it's doing nothing.
IOW, even when your "acpi config" is good, you can save a lot of power. Not minutes, but even hours. How, is detailed in the article.
So, what does it change for an end user looking at which application sucks more power?
Can they fix tha application? No.
Can they stop the application? Don't think so. If they are using it, it means they need it.
before i get screamed the hell out of, i meant machines less than 2.6.2
I don't know who moderated the parent "-1 troll" of course, but I do know that they're idiots. This is very useful for developers and distributors, much less so for average users. Sheesh. Hope karma is a bitch.
I have just tried the thing. I achieved less than 20 wakeups per second when my KDE desktop is idle, but learned a few things on the way. For example, by using a USB mouse instead of the laptop touchpad I am unable to reach state C3. It's reached when I unplug the mouse. I suppose I'll have to put up with it, because I can't stand the touchpad. On the other hand, I used to have KMail opened in a second virtual desktop to check for mail every 60 seconds, but I discovered that the bastard was waking up twice a second for no apparent reason, so I have started to use Korn (the mail check systray thingy). There are still some applications that wake up for no reason apparently. For example, why does klipper wake up once per second? And the same goes for kwrapper. I don't even know what that is. Can somebody explain in detail? Google isn't very specific about it.
:P
But yes, the application is very interesting. Sorry, Intel, my laptop has an AMD processor. The next one will be Intel, with an Intel graphics card and an Intel wireless card. I promise.
"PowerTOP gives a snapshot of what apps are consuming the most power."
Cool, now we'll see things like "New Notepad-lite with reduced power usage!" Maybe "new, less-bloated app" won't be far off. I can dream.
Our FreeBSD servers auto-throttle their CPU speeds down when idle. The average runtime on our monitored UPS has gone from 60 to 75 minutes. Even if electricity were free, and even if air conditioners were free, and even if we didn't care about wasting energy for no good reason, that still means we have 15 more minutes to get the generator up and running in the event of a long power outage.
Maybe that's not much to you, but it's pretty darn nice for us.
Dewey, what part of this looks like authorities should be involved?
The biggest and easiest power savings come from CPU frequency scaling (if your processor supports it). Linux has long done a good pretty good job of putting the CPU to sleep and low power states when it can.
For older Athlon/Duron processors installing/running athcool makes a significant difference in power consumption (as long as it runs stable on your hardware, which it isn't guaranteed to do). On one of our old servers it reduced idle power draw from 100w to 65w.
It might depending on how idle your servers are. The more idle they are, the bigger the possibility for power savings. While they are obviously targeting laptop battery consumption, all Linux machines running this tool (and kernel 2.6.21 or later which has the dynticks feature) will be identify what is waking up their processor from low power states.
Much of the work put into optimizing battery life on laptops will also apply to desktops and servers alike.
ppl who need security run 2.4
because 2.6 isn't as well audited and a moving target. yes, this is a problem. we would like to use 2.6, but we can't. (and auditing is tricky, u can't just pay a bunch of specialists to do it, beside being very expensive, u need several audits from different companies + some time running that kernel in production)
Utter nonsense.
All I've had to do on the last 4 laptops I've tried was to just run the Ubuntu installer/livecd.
Try running a distro that's not from the dark ages. Even Debian seems to fit this description.
A Pirate and a Puritan look the same on a balance sheet.
A KDE developer used it and made a patch for arts on his blog. I look forward to what other developers find and fix.
Do you changes clothes while making the "chee-chee-cha-cha-choh" transformation sound?
Um, what? I run NetBackup Server (version 6) on RHEL 4 (kernel 2.6.9) just fine. I have NBU clients running on RHEL 5 (kernel 2.6.18.) You just need compat-libstdc++-296 and compat-libstdc++-33 installed. It even tells you that in the manual.
No idea if earlier versions of NBU work on 2.6, but v6 is old enough now that it can be considered stable.
So who is going to write the daemon that shuts down various services like beagled or reconfigures running applications (i.e. turns off animated cursors) when running on batteries? Of course, switching everything back on when line power is provided would be a must.
Persistent Volume manager for Kubernetes - https://github.com/dwimsey/openshift-pvmanager
This is not really true. There are still widespread problems for laptop users especially for older laptops that had ACPI quirks. Linux and the laptop vendors never really got ACPI straight for a few years. Also, lets not forget about all the function keys that are not always well supported (though my toshiba portege is fantastic except for ACPI being rubbish which i suspect is the vendors fault). Oh and I have used 2.4 and subsequently 2.6 kernels on it starting around 2004.
ACPI and throttling can save power, nobody denies that. What they're doing here is an additional way of saving power. It is fully compatible with both ACPI and CPU throttling.
;) but I understand that this is definitely not CPU throttling.
At the moment, for example, my Core 2 Duo runs at 1.6 Ghz, for Linux's CPU frequency governor is set to "ondemand" and this Core 2 Duo supports two speed by default (that's an Intel limitation, not a Linux one): 1.6 Ghz or 1.86 Ghz, when the load on the CPU is high.
What they're talking about in the article is another great way to save power: basically it's unnecessary to have interrupts waking up the system when you know nobody while need these.
And it is compatible with already existing method of saving power like, say, CPU throttling.
So, once more, this neither ACPI nor CPU throttling.
Instead of having a 1000 Hz or 250 Hz timer waking up the CPU constantly, if I understand correctly this method switches to a CONFIG_NO_HZ (tickless) idle mode when it knows nothing needs the "regular" tick. I'm no Gleixner (the kernel hacker who's the timer-pro
Last time I hacked to save power was using the HLT instruction on, what, 486? So I'm a bit rusty.
Still, this new kernel parameter an that little util built on top of it will have good implications for battery life, electricity bills, etc.
Anybody commenting on this saying: "nothing to see here, move along" MUST be modded down.
My last 6 power bills add up to $127.10. Of course, that doesn't include the hottest months. I can't view back further than 6 months, or otherwise I'd provide a more exact number. I do know that my power bills don't go up that much during the summer, so I just guesstimated at the $300 value. No doubt it helps that I live in a 1-bedroom apartment and not a house.
Ben Hocking
Need a professional organizer?
Considering it is slackware, very few people are effected by it.
Based off your 220W figure, that results in approximately 158.4 kWh per billing cycle. Yowsa. folding@home would appear to be a good cause, but I agree that it would be nice to know up front how much you're actually "donating" to them. OTOH, if you don't notice an extra 158 kWh on your power bill, then perhaps you're not really going to care that much.
Ben Hocking
Need a professional organizer?
if you've a process eating 100% of the cpu power all the time
1. It's not so much acpi, but cpu frequency scaling I should have mentioned. Sorry, wrong terminology.
2. My point is that the unsexy work of sophisticated uses of frequency scaling would probably help more on a laptop. I'm estimating the most power consumption is the lcd panel followed by the cpu which is where the frequency scaling helps.
3. I run a bunch of servers and a storage array and it would be great if the disks would run at lower power consumption, kind of like frequency scaling at non-peak hours. I know it would cut down on the heat we generate and power/cooling requirements. That's dollars and cents savings that look good. Can hdparm settings lower my power consumption?
http://www.maxineudall.com/2010/02/should-economists-be-sued-for-malpractice.html
Most people i know still run 2.4.x
That's like me saying, "Most people I know run OS/2 and BeOS, so this doesn't help."
Sure it could be true, but so what? I just have a skewed sample and it just means I hang out with weirdos (well.. I do but they don't run OS/2 or BeOS, well.. they don't currently run them).
"If you are going through hell, keep going." - Winston Churchill
Even though I do use AC (although sparingly). My heat comes from the radiators, and hot water comes from our apartment complex. (I.e., it's included in the rent.) Also, my stove/oven uses gas. Just throwing all of that out there since I seem to have generated a little bit of skepticism with my original claims.
Would anyone not notice (other than rich people with mansions, etc.) an increase of 2.5 kWh/day? Even if you use 10x the electricity that I do, that'd be a 10% increase.
Ben Hocking
Need a professional organizer?
On a laptop, it would not only be stupid, but lots of hardware wouldn't even work. Many laptops have the Intel ipw series wifi radios; these have drivers only for 2.6 kernels. I can't imagine anyone is running a recent laptop with a 2.4 kernel.
Penny - plain text accounting
Quo usque tandem abutere, Nimbus, patientia nostra?
gnome-power-manager as the biggest power hog on the system.
Linus Torvalds called timer-related improvements "the big change during 2.6.21.", and the improvement that matters here I think is the Tickless Kernel, that allows you to sleep for more time. The OS no longer needs to be interrupted N times a second (250 times a second usually, 1000 times for multimedia systems, configurable at compilation time (example: CONFIG_HZ_250=y)), since it can be smarter now. Check: This support is not ready for all the architectures yet, but this should happen soon, since the benefit of saving power and having lower temperatures can be huge. And yes, it might be good for your sever farm, too. This is useful even if the system is busy.
What a big load of baloney.
I also disagree with you, I think you missed another requirement. Not dark ages, and not brand spanking new either.
/. the other day? The x800 or somesuch? I bet it doesn't work with the ubuntu installer... yet...
For instance, I bought a machine based on the nvidia 6150 embedded graphics card. It was an absolute nightmare with incomatible, buggy ACPI, the network only worked after a warm boot, sound was horribly distorted with a background whine and sleep only worked once.
That nightmare lasted about three months. By then the kernel developers (with not even a bug report from me) had fixed absolutely everything wrong with it. In another three months, the fixed kernel had made its way into ubuntu and so following your advice would've worked perfectly.
For another example, what was that video card we heard about on
PS: There is also 'terribly made', I find really crap hardware works worse in linux than in windows. For instance, those extremely cheap quickcam thingies which require a software driver to mask gross deficencies in their sensor.
... its a tie between my USB-powered arc welder and all the kewl blue LEDs.
Have gnu, will travel.
Of course, he's no Joe Six-Pack, either. He's definitely where I got most of my nerd genes from. Yeah, I suppose I was thinking too much from my own no-child, low-power consumption perspective, though. For me, it'd literally double my power bill. Trust me, I'd notice.
Ben Hocking
Need a professional organizer?
Aah, well 6 is pretty nice, especially with respect to BMR on Windows. It actually makes me not want to slit my wrists every time I have to migrate a Windows box between hardware. Though I do have to say the support has gone downhill since Symantec took over...
Patrick, is that you?
From the man page of kshell (link):
kwrapper tries to make the application look like it was actually started directly and not via kdeinit. Like kshell it passes application name, arguments, complete environment and current working directory to kdeinit.
Additionally it - tries to redirect application output to the console from which kwrapper was started - waits for the application to finish (but does not return its return value) - passes most signals it gets to the process of the started applicationThe signal passing allows you to use Ctrl-C to break the started application or Ctrl-Z to stop it.
Note: With the use of kwrapper you will have one more process running and also the signal passing and output redirection may not work properly.
So, nobody has made a "Power Bottom" joke yet?
When one app requests to be woken up every five seconds, another app wants to be woken up every two seconds and another app needs to run every second, then why is it necessary to have three wakeup events?
Shouldn't the kernel be able to merge the events? And even if it is currently not possible, it is not rocket science to invent an API that allows the kernel to synchronize these events. In the example above the wakeup rate could be reduced from 1.7 to 1.0 wakeup per second, possibly resulting in 70% improved battery live!
Well, that's not entirely true. Many servers do spend a lot of time idle. Take for instance company internal servers -- they spend HUGE amounts of time idle, about 24 - 8 = 16 hours of each day. OK, subtract one hour for backups. Still, that's a huge amount of power being wasted on an idle machine. Also, there are a lot of web servers that serve country-specific sites. These machines may not be idle enough to spin down their hard drives, but they sure are idle enough to save some serious CPU power. They would probably benefit from shutting down entire CPUs during these "idle periods".