You are not logged in.
i *think* BFS was designed to spam the LKML
hehe, that could be true, but there seems to be something wrong with the backend of lkml.org today, I can't access a lot of the recent archives directly, and doing a search for "bfs molnar" brings up some results, but Molnars email is only showing up in the google cache now ), I blame bfs
====* -- Joke
O
\|/ --- Me
/ \ Whooooosh
Offline
Huh. The old patch just took out everything in CFS (sched.c) and moved it to sched_cfs.c, made a new sched.c and created sched_bfs.c as the BFS scheduler. In other words, a clean approach allowing you to choose between the schedulers. Looks like this approach got old to maintain.
The diffstat (I guess that's what it's called) I pasted was from the 209 patch.
Got Leenucks? :: Arch: Power in simplicity :: Get Counted! Registered Linux User #392717 :: Blog thingy
Offline
Testing it right now and it does seem to make a differnence here. Runnnig HD youtube videos + opening a few tabs in firefox is something that used to reliably produce stutter, with bfs it's perfectly smooth. Still, i'll have to think about some more standardized tests.
thats something i need to see to believe...and i couldnt get X to start with bfs
i'll try again this morning...
Offline
Are you guys using 2.6.30 or the rc9 for 2.6.31?
Offline
Just tried 2.6.31-rc9zen with bfs enabled..... working very nice here for me... think the best choice for a desktop at the moment....
Offline
could you please pastebin your kernel config?
Yah, would be useful...
xclip seems to have disappeared from the repos
EDIT: err... I am using a default config (vanilla sources) and I am still getting that error. I applied the patch on http://www.kernel.org/pub/linux/kernel/ … c9.tar.bz2
Last edited by pyther (2009-09-09 12:44:43)
Offline
bangkok_manouel wrote:could you please pastebin your kernel config?
Yah, would be useful...
xclip seems to have disappeared from the reposEDIT: err... I am using a default config (vanilla sources) and I am still getting that error. I applied the patch on http://www.kernel.org/pub/linux/kernel/ … c9.tar.bz2
the error seems to be related to NUMA, maybe try to disable that if you don't have a NUMA capable CPU. otherwise, you can try those configs, they're arch configs with very very few modifications (following ck's advices on bfs' faq) so they should be pretty much bullet proof whatever the hardware you own.
i686
http://omploader.org/vMmJmMA/config
x86_64
http://omploader.org/vMmJmMw/config.x86_64
Offline
pyther wrote:bangkok_manouel wrote:could you please pastebin your kernel config?
Yah, would be useful...
xclip seems to have disappeared from the reposEDIT: err... I am using a default config (vanilla sources) and I am still getting that error. I applied the patch on http://www.kernel.org/pub/linux/kernel/ … c9.tar.bz2
the error seems to be related to NUMA, maybe try to disable that if you don't have a NUMA capable CPU. otherwise, you can try those configs, they're arch configs with very very few modifications (following ck's advices on bfs' faq) so they should be pretty much bullet proof whatever the hardware you own.
i686
http://omploader.org/vMmJmMA/config
x86_64
http://omploader.org/vMmJmMw/config.x86_64
Yes, bfs isn't compatible with NUMA and dynticks. I don't think its made to spam lklm There's quite interesting discussion going on and it seems mainline scheduler will see some nice tuning :>
Offline
Just tried 2.6.31-rc9zen with bfs enabled..... working very nice here for me... think the best choice for a desktop at the moment....
Can you post your config and I want to add to the PKGBUILD (better way then serching some thing in config) make menu config and I got error
I can't remember it becouse I have to clean src dir
Last edited by SpeedVin (2009-09-09 17:36:42)
Shell Scripter | C/C++/Python/Java Coder | ZSH
Offline
Removing NUMA caused the kernel to compile, but when I tried to boot it I got at traceback. I'm compiling again to see what happens.
If it's worth...
Jens results (however, you can probably easily ignore CFS results here due to some bug in application):
http://marc.info/?l=linux-kernel&m=125249882923867&w=2
Probably correct CFS times:
http://marc.info/?l=linux-kernel&m=125251947426967&w=2
AFAIK they made some tuning to CFS and it should easily beat BFS on desktops. Btw. CFS works completely fine here.
Offline
If anybody has a rooted Android phone. Try out the experimental build of the Cyanogen Mod it uses the bfs and it is a whole heck of a lot better than before.
Right now I am trying out my new PKGBUILD for the patch to see if there is any problems. I remember when Con first got into Kernel Hacking, bringing the realtime patches to the 2.4 kernel at the time made Linux way lot better. Just hope this new PKGBUILD works for everybody.
Offline
Solid1986Snake wrote:Just tried 2.6.31-rc9zen with bfs enabled..... working very nice here for me... think the best choice for a desktop at the moment....
Can you post your config and I want to add to the PKGBUILD (better way then serching some thing in config) make menu config and I got error
I can't remember it becouse I have to clean src dir
Heres my config: http://nopaste.info/ca658f57e3.html
But I changed a lot and its still not clean... it was more a fast try....
Offline
Removing NUMA caused the kernel to compile, but when I tried to boot it I got at traceback. I'm compiling again to see what happens.
You can also try choosing SLAB instead of SLUB... I compiled 2.6.30 with BFS, but it kills my keyboard and mouse when KDM loads. When I mark !KDM I can login to vt, but I can't say a thing about latencies then
Offline
SpeedVin wrote:Solid1986Snake wrote:Just tried 2.6.31-rc9zen with bfs enabled..... working very nice here for me... think the best choice for a desktop at the moment....
Can you post your config and I want to add to the PKGBUILD (better way then serching some thing in config) make menu config and I got error
I can't remember it becouse I have to clean src dirHeres my config: http://nopaste.info/ca658f57e3.html
But I changed a lot and its still not clean... it was more a fast try....
Can you post it somewhere I don't like to delate 2500+ number from lines , and can you post PKGBUILD you build kernel with becouse I see the congig was auto-genrated by menuconfig?
I got problem's with menuconfig option in orginal PKGBUILD
Thanks.
Shell Scripter | C/C++/Python/Java Coder | ZSH
Offline
Hopefully this spurs development on the mainline scheduler and both create an ultimate scheduler of justice !
Offline
Solid1986Snake wrote:SpeedVin wrote:Can you post your config and I want to add to the PKGBUILD (better way then serching some thing in config) make menu config and I got error
I can't remember it becouse I have to clean src dirHeres my config: http://nopaste.info/ca658f57e3.html
But I changed a lot and its still not clean... it was more a fast try....
Can you post it somewhere I don't like to delate 2500+ number from lines , and can you post PKGBUILD you build kernel with becouse I see the congig was auto-genrated by menuconfig?
I got problem's with menuconfig option in orginal PKGBUILD
Thanks.
I used kernel26zen-git PKGBUILD from aur and generated my own config...
I get a lot of wakeups with this scheduler ("Rescheduling Interrupts --- 1000) Somebody else or is it just a misconfigured kernelconfig?
Offline
SpeedVin wrote:Solid1986Snake wrote:Heres my config: http://nopaste.info/ca658f57e3.html
But I changed a lot and its still not clean... it was more a fast try....
Can you post it somewhere I don't like to delate 2500+ number from lines , and can you post PKGBUILD you build kernel with becouse I see the congig was auto-genrated by menuconfig?
I got problem's with menuconfig option in orginal PKGBUILD
Thanks.I used kernel26zen-git PKGBUILD from aur and generated my own config...
I get a lot of wakeups with this scheduler ("Rescheduling Interrupts --- 1000) Somebody else or is it just a misconfigured kernelconfig?
Ok I got my config but when I building I got error:
Failed... Your source tree might be broken. Run 'make mrproper' in src/kernel26zen to clean it up
Source is not broken config is good what wrong?
Shell Scripter | C/C++/Python/Java Coder | ZSH
Offline
Btw. if someone has problems with the mainline scheduler maybe this will help:
# echo NO_NEW_FAIR_SLEEPERS> /debug/sched_features
Btw., NO_NEW_FAIR_SLEEPERS is something that will turn the scheduler
into a more classic fair scheduler (like BFS is too).
Offline
Does anybody know what happens with dynticks + BFS? I wanted to try it out, but I'm on a laptop here.
Offline
Does anybody know what happens with dynticks + BFS? I wanted to try it out, but I'm on a laptop here.
BFS is not compatible with dynticks, but I have no idea what will happen
Last edited by pawels64 (2009-09-10 18:34:31)
Offline
Solid1986Snake wrote:SpeedVin wrote:Can you post it somewhere I don't like to delate 2500+ number from lines , and can you post PKGBUILD you build kernel with becouse I see the congig was auto-genrated by menuconfig?
I got problem's with menuconfig option in orginal PKGBUILD
Thanks.I used kernel26zen-git PKGBUILD from aur and generated my own config...
I get a lot of wakeups with this scheduler ("Rescheduling Interrupts --- 1000) Somebody else or is it just a misconfigured kernelconfig?
Ok I got my config but when I building I got error:
Failed... Your source tree might be broken. Run 'make mrproper' in src/kernel26zen to clean it up
Source is not broken config is good what wrong?
Did you insert your own config file? Go to the src dir of the build/kernelzen and run make menuconfig. Then copy .config file to your root package dir with name configxx.saved ...
And don't use tools like yaourt... they will not work.
Offline
Btw. if someone has problems with the mainline scheduler maybe this will help:
# echo NO_NEW_FAIR_SLEEPERS> /debug/sched_features
Btw., NO_NEW_FAIR_SLEEPERS is something that will turn the scheduler
into a more classic fair scheduler (like BFS is too).
Yes, but there is no /debug in Arch by default. So maybe people have to create de directory and mount debugfs there first. Maybe something like this (as root)?:
# mkdir /debug
# mount -t debugfs debugfs /debug
# cat /debug/sched_features (to test if it worked)
Offline
@Bogart
I have no idea However, this change will be in mainline tree (I don't know if it will be backported). This change should solve latency problems which some people can suffer from. It seems bfs can go to retirement now
(very early
).
here's something about debugfs:
http://www.linuxtopia.org/online_books/ … 09s07.html
mount -t debugfs none /sys/kernel/debug
Offline
SpeedVin wrote:Solid1986Snake wrote:I used kernel26zen-git PKGBUILD from aur and generated my own config...
I get a lot of wakeups with this scheduler ("Rescheduling Interrupts --- 1000) Somebody else or is it just a misconfigured kernelconfig?
Ok I got my config but when I building I got error:
Failed... Your source tree might be broken. Run 'make mrproper' in src/kernel26zen to clean it up
Source is not broken config is good what wrong?
Did you insert your own config file? Go to the src dir of the build/kernelzen and run make menuconfig. Then copy .config file to your root package dir with name configxx.saved ...
And don't use tools like yaourt... they will not work.
Hello.
I insterded my own config and I don't use Yaourt I compile all by abs
When I do make menuconfig in /src dir I got following error:
make: *** No regules to make object `menuconfig'. Stop.
Shell Scripter | C/C++/Python/Java Coder | ZSH
Offline