OESF Portables Forum
General Forums => General Discussion => Topic started by: Da_Blitz on April 20, 2007, 12:26:42 am
-
THere have been a couple of new kernel options that have been introduced latley and i am wondering if any one else knows about them or wants to comment
the first is the "tickless kernel" which does away with "switch tasks every 1000hz" and changes it to "go to sleep until somthing needs the cpu. as we all know doing a hlt or simmilar saves power and it allows for great power savings when put together with other things
the other new one in the next kernel is compleatly fair queing (http://kerneltrap.org/node/8059), sounds intresting to me but from what i can see this will play very nicly with the above tickless kernel. the advantages of this squedular is better responsivness, faster and less relinace on ticks and jiffies (i belive it uses nano second tracking of execution time)
i should mention laptop mode here for compleatness, its been around for awhile but afaik it is very rarely enabled. what it does is only spin up the hard drive when somthing needs to be read OR a ceritn amount of time has elapsed. it stores all the writes and writes them together to help avoid the "power dump" on hard drive spin up and makes the drive idle longer
one of the more obvios ones is cpufreq scaling which explains its self, combined this with policies (cpufreqd anyone, powernowd or others dont really have the same featureset as it can ramp up the speed when you launch mplayer rather on the load as to avoid ossiclations where you go from 100% to 75% and repeat
well those are some of the options, now for the question. has anyone done some power testing with diffrent kernel options? i blew the power supply on my last Z (about to get a new one) but it still works so i was planning to hook up a multimeter with datalogging via rs232 and do some automated powertesting with the dead unit.
any intrest in doing so? what do you want tested. in an ideal world i would love to see what an arch optimised vs processor optimised kernel/userspace does and EABI vs the normal stuff
i dont belive it would be too hard to set it up so the Z is permanttly net connected as well for a dev enviroment or to run your own power testing stuff however i see the timeframe on that one bieng about 2 months
i am aming to punch up the idle load run time by at least 3 hours and the mp3 playing load by 1 hour. moive playback i am aiming for 30 minutes however this one is hard to quanitfy as there are many diffrent moive types that are optimised diffrently for the Z (ie divx, mpeg4, xvid, h263, not to mention container formats)
-
THere have been a couple of new kernel options that have been introduced latley and i am wondering if any one else knows about them or wants to comment
the first is the "tickless kernel" which does away with "switch tasks every 1000hz" and changes it to "go to sleep until somthing needs the cpu. as we all know doing a hlt or simmilar saves power and it allows for great power savings when put together with other things
Don't forget that switching power states also takes power, so entering deep sleep mode 100 times per second takes more power than to just run at full speed.
Dyntick has been around for a while and has been enabled in the openzaurus and angstrom kernels for quite some months now.
the other new one in the next kernel is compleatly fair queing (http://kerneltrap.org/node/8059), sounds intresting to me but from what i can see this will play very nicly with the above tickless kernel. the advantages of this squedular is better responsivness, faster and less relinace on ticks and jiffies (i belive it uses nano second tracking of execution time)
CFS is nice, but is basically 'hype'
i should mention laptop mode here for compleatness, its been around for awhile but afaik it is very rarely enabled. what it does is only spin up the hard drive when somthing needs to be read OR a ceritn amount of time has elapsed. it stores all the writes and writes them together to help avoid the "power dump" on hard drive spin up and makes the drive idle longer
On embedded systems applications are being smart enough to not touch the disk too much (e.g. logging, polling, etc), so 'laptop mode' is for stupid desktop stuff that keeps hitting the disk
one of the more obvios ones is cpufreq scaling which explains its self, combined this with policies (cpufreqd anyone, powernowd or others dont really have the same featureset as it can ramp up the speed when you launch mplayer rather on the load as to avoid ossiclations where you go from 100% to 75% and repeat
Cpufreq doesn't save as much as you'd think, since in most cases it won't to voltage scaling, which is what you really want. And remember, on a pda the backlight is consuming like 80% of the total power, so turning brightness down a notch has more impact than slowing down the cpu and forcing you to having the screen on longer.
well those are some of the options, now for the question. has anyone done some power testing with diffrent kernel options? i blew the power supply on my last Z (about to get a new one) but it still works so i was planning to hook up a multimeter with datalogging via rs232 and do some automated powertesting with the dead unit.
Like I indicated above, you have to know when to apply such features, because you can end up using more power by enabling conflicting options.
any intrest in doing so? what do you want tested. in an ideal world i would love to see what an arch optimised vs processor optimised kernel/userspace does and EABI vs the normal stuff
Fixing apps to stop polling and thrashing the disk would be of more use than using 'gcc -OMG'
i dont belive it would be too hard to set it up so the Z is permanttly net connected as well for a dev enviroment or to run your own power testing stuff however i see the timeframe on that one bieng about 2 months
turning down your brightness is still the best way to save power. For the rest, use software that has been fixed for 'embedded' usage, we have a shitload of patches in OE that remove polling and disk access for various apps.
-
thats intresting and exacttly what i was hoping for
hahah, i dont know anyone who would change the cpu speed 100 times a seccond, thats what halt is for. i find that depending on usage every 2 seconds is neough, or if the deamon supports it, 0.5 secs on ramp up and 2 secs on ramp down to maintain responivness. (YMMV, personal opinion)
i think at the very least i would like to see the diffrence made by changing the backlight settings, when listining to mp3's i would have the display off but would still like to get a bit more juice out of the battries, btw is that with the pxa at typical or full load as i was under the impresion that when the pxa ramps up to 100% it tends to drain the battries a bit quickly (and has been my experince)
the -OMG is more intresting for media players and thats about it, must think a bit more before saying stuff like that as my mind is the right context of what i say, and in many cases i dont give enugh info
i was unedr the impresion that the pxa2xx cpu freq did volt scaling as well, if not i have seen some code that has at least mentioned that stuff, and as its so tightly tied to cpufreq you would think that it would be together
as for OZ stuff as well as armtstrong i havent been following it scince my Z died, i suppose i should subscribe up to the armstrong mailing list and catch up
as for CFS, it may be hype but is it all hype or is there some performance gain that could be useful (perhaps not just in power saving)