14:02:48 <jskarvad> #startmeeting PowerManagement
14:02:48 <zodbot> Meeting started Wed Feb 23 14:02:48 2011 UTC.  The chair is jskarvad. Information about MeetBot at http://wiki.debian.org/MeetBot.
14:02:48 <zodbot> Useful Commands: #action #agreed #halp #info #idea #link #topic.
14:04:06 <jskarvad> hi all
14:04:35 <jskarvad> I digged into tuned.
14:05:16 <jskarvad> I created experimental patch that incorporates dynamic EEEPC FSB switching into CPU tuning plugin.
14:05:23 <HanzZ> I've checkouted it yesterday :)
14:06:20 <jskarvad> HanzZ: it is experimental in my local branch and not yet commited ;)
14:06:53 <jskarvad> to server
14:07:12 <HanzZ> I see :) I meant I've checked out tuned :-P
14:07:37 <jskarvad> HanzZ: good start :)
14:08:26 <jmarko> and how big effect it has?
14:09:20 <jskarvad> Purpose of the patch: the FSB switch takes too long, so it is not handled automatically by kernel and have to be handled in user space
14:10:03 <jskarvad> jmarko: currently we have static FSB tuning in F14 tuned, if you enable tuned profile laptop-battery-powersave the FSB is statically reduced
14:10:29 <jskarvad> jmarko: according to our tests it is about 1.5 W savings in active idle and more on loads
14:10:38 <HanzZ> great :)
14:12:27 <jskarvad> jmarko: the purpose of the patch is to make it dynamic, because decreasing FSB also decreases computational power, so the same task will be computed longer with lower FSB and then there is no energy savings
14:13:31 <jskarvad> In the proposed patch the FSB frequency is switched dynamically according to system load
14:14:41 <jskarvad> The functionality can be controlled by the tuned CPU plugin config directive 'eeefsb=True/False' and by default I enabled it for laptop-battery-powersave
14:14:52 <jskarvad> profile
14:15:12 <jskarvad> Current implementation:
14:15:14 <jmarko> so it is only 2 step switching (power save / fast)
14:16:24 <jskarvad> jmarko: Currently the EEEPC supports three modes: powersave, high performance (normal), and super high performance (slightly overclocked), but we only use normal and powersave
14:20:44 <jskarvad> The tuned implementation can be used generally, but currently we have only backend code for the EEEPC FSB switching
14:20:44 <jmarko> what about do a banchmark of all tree states (power-save/high performance/super high performance)
14:21:40 <jskarvad> jmarko: So far I did not benchmark super high performance
14:22:05 <jskarvad> jmarko: I will check this
14:22:30 <jskarvad> jmarko: check our wiki, I am going to create page with benchmark results
14:24:23 <jskarvad> About the implementation: currently the FSB switch can take place in every 10 secs.
14:24:41 <jmarko> can you send me a link to the wiki page?
14:25:15 <jskarvad> jmarko: sure, SIG wiki: https://fedoraproject.org/wiki/SIGs/PowerManagement
14:25:58 <jskarvad> jmarko: there are also some older results, but I am going to update it
14:26:19 <jmarko> thanks
14:27:14 <jskarvad> The FSB is switched to lower frequency if the average system load is bellow 20 percent and it switches back to normal if the average load is above 40 percent.
14:27:46 <jskarvad> The small hysteresis was added to eliminate periodic switching.
14:28:11 <jskarvad> The coefficients were taken by hand, but I am experimenting with various settings.
14:28:58 <jskarvad> Currently I am testing this feature and if satisfied I will post the patch to jvcelak for review / inclusion, becouse I would like to have this in f15.
14:32:49 <jskarvad> Next thing: I also overviewed the PM-QoS latency settings in tuned.
14:33:03 <jmarko> It can by cool write script which run test with different % setting and then report the results for all possible combinations (with step 1%)
14:33:51 <jskarvad> jmarko: Good idea, but it also depends on type of load
14:34:30 <jskarvad> jmarko: so the solution space of this task is huge
14:35:13 <jskarvad> jmarko: but I will take it into account
14:35:38 <jmarko> I can help with the script
14:40:42 <jskarvad> jmarko: we can use it to get limits for switching for specific machine, e.g. when the switch will save energy
14:41:14 <jskarvad> jskarvad: we would need to manipulate the cpu load
14:41:53 <jskarvad> jmarko: so this approach doesn't help with the hysterisi
14:42:11 <jskarvad> jmarko: for the hysteresis setting we would need 'real life' examples
14:42:37 <jmarko> ok so we need to create one real live banchmark
14:42:45 <HanzZ> when we're in that brainstorming mood, what about create some real life benchmark
14:42:57 <HanzZ> using for example some firefox API to simulate browsing
14:43:04 <HanzZ> just idea ;)
14:43:24 <jskarvad> HanzZ: +1 for OS independent real life benchmark
14:43:39 <HanzZ> let it to check youtube pages, play music in amarok from time to time
14:43:52 <jskarvad> HanzZ: I am looking for such thing
14:44:07 <jskarvad> HanzZ: feel free to prepare one, it would help us
14:44:07 <HanzZ> *amarok == some player, that's what I use...
14:44:09 <jmarko> but it will be hard to do it os independent
14:44:17 <HanzZ> well, I'm ok with linux so far
14:44:26 <HanzZ> with firefox it would not be so hard to do it independed
14:44:35 <HanzZ> *independed
14:44:59 <jskarvad> Hanzz: It would be good to be able to do inter OS comparisons
14:45:11 <HanzZ> yeah, I know your motivation here
14:45:11 <jmarko> yes but you need to test creating documents, reading pdf, reading email, listen to music, watch videos
14:45:16 <jskarvad> HanzZ: According to my personal tests Fedora doesn't stay bad
14:45:52 <HanzZ> jmarko: that can't be done independently, but you can at least do *some* things
14:46:03 <jmarko> ok creating documents in libre office can be os independent
14:46:18 <jmarko> using ff and Thunderbird too
14:46:23 <jskarvad> HanzZ: Feel free to start such project, btw I will add it to our todo page
14:47:00 <jmarko> what about vlc for music and video?
14:47:01 <jskarvad> currently I am looking for the suitable, reliable and inter-platform capable free benchmark which can be used for the power consumption measurement
14:47:02 <HanzZ> jmarko: you can have special OS-depend modules anyway which would use some other apps. The idea is to have some common API for invocation of those scripts
14:47:26 <HanzZ> *scripts == modules
14:47:54 <HanzZ> well, I think I can help here. At least I can start with "surfing the internet" benchmark
14:47:54 <jskarvad> HanzZ: currently bltk have some loads, but we need something more robust
14:48:06 <jskarvad> HanzZ: great
14:48:06 <HanzZ> jskarvad: have to google it for more info
14:48:22 <jskarvad> HanzZ: +1 from me
14:49:52 <jskarvad> I really appreciated any help with this
14:50:54 <jskarvad> It would be good not to have only desktop tests but also server tests, e.g. requests to HTTP server, etc.
14:51:53 <jskarvad> HanzZ: counting with you for PM ;)
14:52:53 <HanzZ> jskarvad: I think there are some statistics SW for webservers
14:53:10 <HanzZ> which simulates "normal" load
14:53:22 <HanzZ> the problem is only with "normal" desktop usage
14:53:33 <fenrus02> ab?
14:53:49 <jmarko> But is the http server a standard use-case of eeepc?
14:54:17 <jskarvad> Any tips would be appreciated, currently I am gathering and incorporating free tests to my personal testsuite
14:54:31 <jskarvad> fenrus02: thanks for tip
14:54:36 <HanzZ> we could even do some logger app to analyze how people work with computer and store their load using sysstat or whatever
14:54:43 <HanzZ> to find out what's "normal"
14:55:02 <jskarvad> HanzZ: this sounds to me like big brother feature ;)
14:55:16 <jmarko> HanzZ: Cool spy feature
14:55:30 <HanzZ> I don' t want to use them secretly :-P
14:55:38 <HanzZ> *them to use it
14:56:38 <jskarvad> anybody interested please sign up to PM SIG mailing list: power-management@lists.fedoraproject.org
14:56:38 <jmarko> Yes you will give there a button: Yes a will be spy-ed by HanzZ
14:58:08 <jskarvad> https://admin.fedoraproject.org/mailman/listinfo/power-management
14:58:22 <jskarvad> And any more tips are welcome.
15:00:16 <jskarvad> OK, according to wiki, there is schedule for another group meeting on 15:00, so I will end it for today
15:00:34 <jskarvad> Thanks all for tips
15:00:39 <HanzZ> :)
15:01:13 <jskarvad> Let's meet again next week.
15:01:29 <jmarko> ok by
15:01:40 <jskarvad> #endmeeting