new exciting features in BFS (making desktop interactive at any load)

broch · 2010-10-05 13:28:28

here is more in-depth information:
http://ck-hack.blogspot.com/2010/10/int … -load.html
and patch available:
http://ck.kolivas.org/patches/bfs/2.6.3 … tity.patch

in addition -zen (git) is patched already with bfs 357. I have no idea if current zen in AUR is updated already or not... nope (quick check in AUR and it looks like out od date)

SanskritFritz · 2010-10-05 14:16:15

Wow, thanks for the news. I will try this asap, BFS is my favorite scheduler so far already, this is a welcome change towards enhancing desktop experience.

archman-cro · 2010-10-05 14:19:49

Thanks for the news, yes! Gonna go build my kernel with it and test to see if it has any real world improvements on my lappie.

* goes to edit the PKGBUILD

broch · 2010-10-05 15:28:06

sorry, wrong thread

Last edited by broch (2010-10-05 15:28:58)

benob · 2010-10-05 16:18:10

I tried it (patching linux26-bfs from aur) and it does not make a large difference when not using a swap (make -j 128 leads to a laggy but usable system). My guess is that the load doesn't go above 20 as make doesn't have enough memory for running more jobs. I will try with a swap to see how it runs.

broch · 2010-10-05 16:20:25

benob wrote:

I tried it (patching linux26-bfs from aur) and it does not make a large difference when not using a swap (make -j 128 leads to a laggy but usable system). My guess is that the load doesn't go above 20 as make doesn't have enough memory for running more jobs. I will try with a swap to see how it runs.

read -ck blog: currently you are limited by the amount of memory available. Do you have that much RAM?

Indeed I was only planning on getting the thread group parent but I managed to use the program group. Bear in mind that the limit for job numbers with make -jX will be the amount of ram you have where you need about 1GB ram per 15 jobs so you'll end up getting ram swappage slowing you down beyond that. I'll be working on a slightly improved version soon.

not using swap? O.K. that is against basic VM theory (kernel needs swap irrelevant of RAM amount). In other words, quite irrelevant.
kernel26-bfs was just patched with bfs-357 minutes ago so it is ready for testing (though I would tweak config per machine)

Last edited by broch (2010-10-05 16:28:46)

benob · 2010-10-05 20:02:58

broch wrote:

read -ck blog: currently you are limited by the amount of memory available. Do you have that much RAM?
not using swap? O.K. that is against basic VM theory (kernel needs swap irrelevant of RAM amount). In other words, quite irrelevant.
kernel26-bfs was just patched with bfs-357 minutes ago so it is ready for testing (though I would tweak config per machine)

You are right about memory. 4G don't go very far. I tried with 15G of swap with the two schedulers (the default and bfs+patch). I ran make -j 128 in the linux tree and then tried to open a browser and check a few web pages. The bottom line is that it's no better with one than the other. The default scheduler gives me nice pointer moves but almost no response as soon as the load reaches 100. BFS make the mouse more jumpy and it's hard to click on anything. But once launched, tasks seem to get some cpu and I could login to my email account.

The real difference was when pressing control-c to interrupt make. It took about 10 secs on the standard scheduler while it was immediate with BFS.

Do you guys get similar results? I am worried that some laptop mode tuning stuff screwed up the benefits (I was on AC though).

broch · 2010-10-05 21:58:22

Do you guys get similar results? I am worried that some laptop mode tuning stuff screwed up the benefits (I was on AC though).

For some time now I am only laptop user (HP dv5000 and netbook with atom 450)

I hope to get better interactivity on the netbook (which I will test as soon I I will compile bfs357). On the laptop I alway have mixed load of CPU and I/O and this worked well even before with "old" BFS after tweaking fs, I/O syscals. I don't have anything that would test CPU load only.
I think though that best extreme testing would be done on some workstation rather than laptop.

Remember though that this is first release. Maybe post your observations here (with detailed system info and configs)
http://ck-hack.blogspot.com/2010/10/int … -load.html
this will definitely help get BFS even better.

sand_man · 2010-10-05 23:09:42

Does anyone know if BFS is a good replacement for -realtime patch for low latency audio production?

ngoonee · 2010-10-06 00:02:34

sand_man wrote:

Does anyone know if BFS is a good replacement for -realtime patch for low latency audio production?

Its adequate, in that it allows lower-latencies without xruns than the standard kernel.

This is during audio-only though, I have not tested BFS with other apps running at the same time eating CPU. If -rt gives you 2 ms, BFS might give you 4 (just a random gauge of how 'low' you can go).

bangkok_manouel · 2010-10-06 00:46:35

sand_man wrote:

Does anyone know if BFS is a good replacement for -realtime patch for low latency audio production?

http://ck-hack.blogspot.com/2010/10/bfs … -time.html

sand_man · 2010-10-06 01:22:01

Thanks guys

broch · 2010-10-06 01:36:43

sand_man wrote:

Does anyone know if BFS is a good replacement for -realtime patch for low latency audio production?

no,
these are completely different bests:
one is designed to make sure that single app (e.g. audio editor) will get all CPU time needed to ensure high quality sound.
On the other hand interactive desktop will balance all threads so under heavy load everybody will get its share: no jumpy mouse or slow window redraw. This is different philosophy, purpose (single vs many balanced).

I would not mix these two. But you can build two kernels -rt and bfs based. One (bfs) for interactive desktop), the other for audio editing.

there is kernel26 in AUR that mixes rt and bfs. It does not make much sense either from desktop interactivity or real time use stand point but it proves that with some effort you can mix anything as long as harm is not done.

Last edited by broch (2010-10-06 12:22:26)

akurei · 2010-10-07 14:33:59

Using kernel26zen-git.
What I'm doing now: Building kernel26zen-git (again), qt, kdelibs, kdebase-workspace, kdeplasma-applets-kdenetworkmanager via "yaourt -Sb". All simultaneously.

top - 16:27:15 up 38 min,  2 users,  load average: 8.17, 8.44, 7.42

Desktop is reacting as if there's no load at all. I love it!

Last edited by akurei (2010-10-07 14:36:25)

SanskritFritz · 2010-10-08 14:58:39

akurei wrote:

Using kernel26zen-git.
What I'm doing now: Building kernel26zen-git (again), qt, kdelibs, kdebase-workspace, kdeplasma-applets-kdenetworkmanager via "yaourt -Sb". All simultaneously.
top - 16:27:15 up 38 min,  2 users,  load average: 8.17, 8.44, 7.42
Desktop is reacting as if there's no load at all. I love it!

Are you sure the new BFS 357 is included? I compiled the kernel from the AUR, but sched.h says
printk(KERN_INFO"BFS CPU scheduler v0.323 by Con Kolivas.\n");

tomk · 2010-10-08 16:50:51

The blog post linked by the OP clearly says

Apply to BFS 357 (3rd revision patch, updated)

which IMO means that you need to apply bfs357 AND the new patch, in that order. 357 on its own does not include the schedule_groups_as_entity patch that motivated the OP to post in the first place.

For anyone using the -pf patchset, like me, -pf11 includes bfs357, and schedule_groups_as_entity applies cleanly to that.

archman-cro · 2010-10-08 17:19:51

@tomk, thanks for demistifying it for me. I thought it would be enough to just apply the latest patch. So it seems I need 4 patches or so. Cool.

Lothium · 2010-10-08 18:00:59

And the kernel26zen-git also includes bfs357 and the new patch, as you can see here: http://git.zen-kernel.org/zen-stable/log/ .

Best regards!

broch · 2010-10-08 18:05:29

archman-cro wrote:

@tomk, thanks for demistifying it for me. I thought it would be enough to just apply the latest patch. So it seems I need 4 patches or so. Cool.

no you don't
CK provides two patches:
1. upgrade (from previous)
http://ck.kolivas.org/patches/bfs/bfs35 … ty-1.patch
2. on clean

2.6.35.7ish

http://ck.kolivas.org/patches/bfs/2.6.3 … ty-1.patch

Last edited by broch (2010-10-08 18:07:43)

archman-cro · 2010-10-08 18:10:44

broch, thanks for replying, but I still don't understand. The patches you linked are outdated, since he posted a new patch today. Is there a patch which includes all the patch parts he releases every day?

graysky · 2010-10-08 18:50:44

Yeah, the blog got updated. There is a single patch now that does it all. I have updated the kernel-ck in the AUR to optionally apply this patch. I just built it, installed, and rebooted. It is truly amazing. I have firefox (open with 7 tabs) and vlc playing a flac file and running 'make -j256 bzImage' in the 2.6.35.7 linux source on /dev/shm/temp and everything is smooth as if my CPU were idle (quad core xeon).

Last edited by graysky (2010-10-08 18:51:51)

archman-cro · 2010-10-08 18:57:14

graysky, "2.6.35.7-sched-bfs357+schedule_groups_as_entity-1.patch" is pushed on Oct 6. Today is Oct 8 and by now there are two new patches out. Or is that file somehow updated but the release date is left to Oct 6? *still very confused*

graysky · 2010-10-08 19:00:09

I saw them, read his blog. One needs major testing, the other I dunno. No, the DTS indicates when they were released.

http://ck-hack.blogspot.com/

Last edited by graysky (2010-10-08 19:00:58)

archman-cro · 2010-10-08 19:11:05

I've just come to a conclusion that the last two (Oct 7th and 8th) releases are for linux 2.6.36. The last one for 2.6.35 is "2.6.35.7-sched-bfs357+schedule_groups_as_entity-1.patch".
Cheers.

graysky · 2010-10-08 19:46:13

archman-cro wrote:

I've just come to a conclusion that the last two (Oct 7th and 8th) releases are for linux 2.6.36. The last one for 2.6.35 is "2.6.35.7-sched-bfs357+schedule_groups_as_entity-1.patch".
Cheers.

Not entirely...

CK wrote:

An updated patch for 2.6.36-rc7-ck1 follows, though it should apply to a BFS357 patched kernel with offsets

Last edited by graysky (2010-10-08 19:46:32)

Arch Linux

#1 2010-10-05 13:28:28

new exciting features in BFS (making desktop interactive at any load)

#2 2010-10-05 14:16:15

Re: new exciting features in BFS (making desktop interactive at any load)

#3 2010-10-05 14:19:49

Re: new exciting features in BFS (making desktop interactive at any load)

#4 2010-10-05 15:28:06

Re: new exciting features in BFS (making desktop interactive at any load)

#5 2010-10-05 16:18:10

Re: new exciting features in BFS (making desktop interactive at any load)

#6 2010-10-05 16:20:25

Re: new exciting features in BFS (making desktop interactive at any load)

#7 2010-10-05 20:02:58

Re: new exciting features in BFS (making desktop interactive at any load)

#8 2010-10-05 21:58:22

Re: new exciting features in BFS (making desktop interactive at any load)

#9 2010-10-05 23:09:42

Re: new exciting features in BFS (making desktop interactive at any load)

#10 2010-10-06 00:02:34

Re: new exciting features in BFS (making desktop interactive at any load)

#11 2010-10-06 00:46:35

Re: new exciting features in BFS (making desktop interactive at any load)

#12 2010-10-06 01:22:01

Re: new exciting features in BFS (making desktop interactive at any load)

#13 2010-10-06 01:36:43

Re: new exciting features in BFS (making desktop interactive at any load)

#14 2010-10-07 14:33:59

Re: new exciting features in BFS (making desktop interactive at any load)

#15 2010-10-08 14:58:39

Re: new exciting features in BFS (making desktop interactive at any load)

#16 2010-10-08 16:50:51

Re: new exciting features in BFS (making desktop interactive at any load)

#17 2010-10-08 17:19:51

Re: new exciting features in BFS (making desktop interactive at any load)

#18 2010-10-08 18:00:59

Re: new exciting features in BFS (making desktop interactive at any load)

#19 2010-10-08 18:05:29

Re: new exciting features in BFS (making desktop interactive at any load)

#20 2010-10-08 18:10:44

Re: new exciting features in BFS (making desktop interactive at any load)

#21 2010-10-08 18:50:44

Re: new exciting features in BFS (making desktop interactive at any load)

#22 2010-10-08 18:57:14

Re: new exciting features in BFS (making desktop interactive at any load)

#23 2010-10-08 19:00:09

Re: new exciting features in BFS (making desktop interactive at any load)

#24 2010-10-08 19:11:05

Re: new exciting features in BFS (making desktop interactive at any load)

#25 2010-10-08 19:46:13

Re: new exciting features in BFS (making desktop interactive at any load)

Board footer