maemo.org - Talk - BFS for the power kernel

maemo.org - Talk (https://talk.maemo.org/index.php)

- Maemo 5 / Fremantle (https://talk.maemo.org/forumdisplay.php?f=40)

- - BFS for the power kernel (https://talk.maemo.org/showthread.php?t=58780)

Re: BFS for the power kernel

Started noticing some issues with freezing and bad priority allocating when using BFS. What is this attributed to? I'm guessing the dropping of cgroups? What kernel tunes actually compliment BFS?

I have noticed a significant decrease in lag and "stuttering" as uptime increases, when compared to CFS.

Re: BFS for the power kernel

That's a good question, unfortunately not one I know the answer to. However, I have included this patch since the posted debs, which may help. I'm planning to post an updated version later today - still working on compcache at the moment, it's a little more work than I thought as it's very different compared to zram (which uses a nice sysfs interface).

You may also try (from Dennis - I haven't tested it yet) the following:

Quote:

Disabling ohmd cgroup module completely (mv /usr/lib/ohm/libohm_cgroups.so /usr/lib/ohm/libohm_cgroups.so_) will cause /syspart to not get mounted at all, thus completely disabling all resource distribution rules

Finally there is schedtool, which has been compiled for the N900 (I think it's been posted previously on this thread?), and can be used within a startup script. Mine has

Code:

start on started hildon-desktop

stop on starting shutdown

console none

service



script



sleep 20

/usr/bin/schedtool -I `pidof hildon-desktop`

/usr/bin/schedtool -D `pidof trackerd`

/usr/bin/schedtool -D `pidof tracker-indexer`

/usr/bin/schedtool -I `pidof mafw-dbus-wrapper`

/usr/bin/schedtool -I `pidof pulseaudio`



end script

Re: BFS for the power kernel

I will have a go with schedtool tasking. Thanks for the post, as well as providing the kernels and modules. Really appreciated time/cycle saver :)

Re: BFS for the power kernel

Code:

    Disabling ohmd cgroup module completely (mv /usr/lib/ohm/libohm_cgroups.so /usr/lib/ohm/libohm_cgroups.so_) will cause /syspart to not get mounted at all, thus completely disabling all resource distribution rules

This makes my device unusable. I did tried it only once but it is the same with hw-vsync enabled.

Re: BFS for the power kernel

yeah i compile schedtool (its good) but i also did chrt (in util-linux) which im using right now(i find it easyer). if you give Xorg policy:SCHED_FIFO and priority:99 i notice better performance. (i messing with some others too but on this process i notice improvement.)

here's a good read http://ck.kolivas.org/patches/bfs/bfs-faq.txt

its also nice to having the device run at 1000 MHz

so far no freezing on my end, not sure if its because my watchdogs are down.

Re: BFS for the power kernel

My device absolutely didnt feel faster after installing this kernel, maybe ive done something wrong.

And another thing, in Opera 11 when u tap the red O the pop-up is all buggy which wasnt in the original kernel

Sorry if my english is bad

Re: BFS for the power kernel

Quote:

Originally Posted by Schriek (Post 979175)

In terminal you can type uname -a or cat /proc/version to see what kernel you are using. If you are right then you should see the suffix -bfs attached to the kernel-name.

Re: BFS for the power kernel

Quote:

Originally Posted by epitaph (Post 979196)

In terminal you can type uname -a or cat /proc/version to see what kernel you are using. If you are right then you should see the suffix -bfs attached to the kernel-name.

uname -a returned 2.6.28-bfs5, so thats ok

Maybe i just expect to much:rolleyes:

Re: BFS for the power kernel

Quote:

Originally Posted by hawaii (Post 978910)

Started noticing some issues with freezing and bad priority allocating when using BFS. What is this attributed to? I'm guessing the dropping of cgroups? What kernel tunes actually compliment BFS?

Hmm, I haven't noticed any freezing myself. Try to see if there is a spike in CPU usage when you experience a freeze. Also, could you provide an example of the bad priority allocation?

BFS doesn't got much to tune. From the FAQ: "The only tunable for the
scheduler itself is the rr_interval value (see documentation)". The same FAQ states you won't have to tune BFS virtually ever :)

Quote:

Originally Posted by Tigerite (Post 979007)

You may also try (from Dennis - I haven't tested it yet) the following:

If I may add my reasoning to do that ;):
BFS doesn't support cgroups and I've read somewhere that some important applications are mlocked by default anyway. Because of this, I don't think the ohmd module has that much effect anymore. I've been running my device without /syspart mounted for a few months already and hadn't experienced any problems.

Quote:

Originally Posted by Tigerite (Post 979007)

Finally there is schedtool, which has been compiled for the N900 (I think it's been posted previously on this thread?), and can be used within a startup script.

Here it is. I'll upload it to the kernel-bfs garage page later so that it'll be easier to find.

Quote:

Originally Posted by Schriek (Post 979175)

And another thing, in Opera 11 when u tap the red O the pop-up is all buggy which wasnt in the original kernel

Some other people in this thread noticed strange behavior using Opera too, see http://talk.maemo.org/showpost.php?p...9&postcount=19 and http://talk.maemo.org/showpost.php?p...5&postcount=31. Opera seems to be the only application affected though.

Quote:

Originally Posted by Schriek (Post 979175)

My device absolutely didnt feel faster after installing this kernel, maybe ive done something wrong.

Quote:

Originally Posted by Schriek (Post 979208)

uname -a returned 2.6.28-bfs5, so thats ok

Maybe i just expect to much:rolleyes:

Using a different CPU scheduler only changes the way CPU time slices are distributed amongst the running processes; in the end your device has still got the same amount of them. When idle, you won't nice much difference with different schedulers as there are plenty of slices to go around anyway. However, at load there are the gains to be found; at least when talking about responsiveness. Try running the kernel for a little longer :)

Re: BFS for the power kernel

Hard locks occurs very randomly and I've yet to experience it in such a manner when using CFS, but I am able to lock up the device on CFS, in any fashion.

With BFS, sometimes it'll lock up when it's under medium load, other times under no load whatsoever. I'll open the status menu, it'll completely halt. Playing a game, reject an incoming call and it'll be hosed. Both the power key and the light sensor (controlling keyboard LED) are unresponsive and I have to battery pull.