omg! Con Kolivas is back!

Ranguvar · 2009-09-07 03:15:59

B wrote:

I just built my kernel with the patch mentioned above, setting the clock till when I see my first system instability .
Edit: launched Transmission and within seconds it seems to have locked up. It has its own process but no window visible, and I can't kill it (SIGTERM or SIGKILL), not even as root...
I think i'm going back to the non-CK kernel scheduler for the time being .

Looks like the same bug you had was replicated and fixed
Check http://ck.kolivas.org/patches/bfs/clone … read.patch, the description sounds very like what you were talking about. An "unkillable" process...
That patch is now in kernel26-bfs.

Allan · 2009-09-07 04:19:23

Hmmm.... I am interested to see the outcome of this: http://lwn.net/Articles/351058/

bangkok_manouel · 2009-09-07 04:22:30

Ranguvar wrote:

B wrote:
I just built my kernel with the patch mentioned above, setting the clock till when I see my first system instability .
Edit: launched Transmission and within seconds it seems to have locked up. It has its own process but no window visible, and I can't kill it (SIGTERM or SIGKILL), not even as root...
I think i'm going back to the non-CK kernel scheduler for the time being .
Looks like the same bug you had was replicated and fixed
Check http://ck.kolivas.org/patches/bfs/clone … read.patch, the description sounds very like what you were talking about. An "unkillable" process...
That patch is now in kernel26-bfs.

what happened to your wolf's eyes ?

oztrailrider · 2009-09-07 04:40:01

Allan wrote:

Hmmm.... I am interested to see the outcome of this: http://lwn.net/Articles/351058/

CK's replied.. http://thread.gmane.org/gmane.linux.kernel/886319

sand_man · 2009-09-07 06:13:13

Wow I've never heard of it before. I'm going to try it out as soon as I get home

Ranguvar · 2009-09-07 06:16:41

bangkok_manouel wrote:

Ranguvar wrote:
B wrote:
I just built my kernel with the patch mentioned above, setting the clock till when I see my first system instability .
Edit: launched Transmission and within seconds it seems to have locked up. It has its own process but no window visible, and I can't kill it (SIGTERM or SIGKILL), not even as root...
I think i'm going back to the non-CK kernel scheduler for the time being .
Looks like the same bug you had was replicated and fixed
Check http://ck.kolivas.org/patches/bfs/clone … read.patch, the description sounds very like what you were talking about. An "unkillable" process...
That patch is now in kernel26-bfs.
what happened to your wolf's eyes ?

They have failed me x.x

Regarding the LKML thread, Con's being a jerk but I can see his point... for me, the real test will be whether this can solve the PC-turns-into-a-glacier-at-first-sign-of-heavy-I/O problem me and quite a few other people have been experiencing. That's all I really care about, so long as throughput isn't _completely_ obliterated.

Last edited by Ranguvar (2009-09-07 06:18:44)

mangus · 2009-09-07 06:29:28

oztrailrider wrote:

Allan wrote:
Hmmm.... I am interested to see the outcome of this: http://lwn.net/Articles/351058/
CK's replied.. http://thread.gmane.org/gmane.linux.kernel/886319

Hero!! LOL! The fight is Back , too...

tomk · 2009-09-07 08:27:44

Con is being Con, and Ingo is being Ingo. Nothing's changed since their last spat - some people will say "Con FTW!!!", others will say "Ingo ROX!!!"

Personally, I think Con should not have responded, as he's already made it clear that he has no intention of going near mainline with this. I really hope he leaves it at that, and lets his code make whatever point he needs to make.

shining · 2009-09-07 08:58:48

tomk wrote:

Con is being Con, and Ingo is being Ingo. Nothing's changed since their last spat - some people will say "Con FTW!!!", others will say "Ingo ROX!!!"
Personally, I think Con should not have responded, as he's already made it clear that he has no intention of going near mainline with this. I really hope he leaves it at that, and lets his code make whatever point he needs to make.

The interaction between both was very stupid. There are clearly cases where Con (BFS) will win, and others where Ingo (CFS) will win.
So there is not one better than the other.
The only way to solve this would have been pluggable cpu sched. But Ingo completely denied that (http://kerneltrap.org/node/4111)

sand_man · 2009-09-07 09:16:59

Just booted 2.6.30-bfs
So far so good. I can't notice much difference yet but it's only been a few minutes

BTW: Compiling a kernel sucks! Takes too long..

bangkok_manouel · 2009-09-07 09:25:07

also, try using schedtool as con advises in the faq. i didn't benchmark anything this time so placebo is a possibility but damn... "i believe i can flyyyyyyyyyyyyyy" playing 1080p easily with my poor man's integrated intel card...

zodmaner · 2009-09-07 09:31:54

bangkok_manouel wrote:

also, try using schedtool as con advises in the faq. i didn't benchmark anything this time so placebo is a possibility but damn... "i believe i can flyyyyyyyyyyyyyy" playing 1080p easily with my poor man's integrated intel card...

Damn, that sounds so tempting... I'm too busy/lazy to compile and maintain another customized kernel, but with that comment perhaps I will give bfs a try.

shining · 2009-09-07 09:52:41

sand_man wrote:

BTW: Compiling a kernel sucks! Takes too long..

The config used can make a HUGE difference. Unless you already enabled only the hw you need.
iirc it takes less than 10min on my laptop (core 2 duo)

sand_man · 2009-09-07 10:13:32

Makes a HUGE difference in nexuiz. It always did run quite well for me but sometimes when the map loads it would freeze up but with this new scheduler, I still had firefox running in the background and it seemed to load up a lot faster. Definitely not placebo I swear!!!

sand_man · 2009-09-07 10:16:04

shining wrote:

sand_man wrote:
BTW: Compiling a kernel sucks! Takes too long..
The config used can make a HUGE difference. Unless you already enabled only the hw you need.
iirc it takes less than 10min on my laptop (core 2 duo)

Yeah core2 duo here too. I've never played with the config before. I just wanted to try this scheduler.

bangkok_manouel · 2009-09-07 10:22:11

[OT] while you guys are recompiling kernels, take the opportunity to include the BFQ patches:
.30 patches http://feanor.sssup.it/~fabio/linux/bfq/patches/
benchmarks http://feanor.sssup.it/~fabio/linux/bfq/results.php [/OT]

.:B:. · 2009-09-07 10:40:09

Ranguvar: you already found out I already did .

sand_man wrote:

Just booted 2.6.30-bfs
So far so good. I can't notice much difference yet but it's only been a few minutes
BTW: Compiling a kernel sucks! Takes too long..

Only if you build a full kernel >:-)

mikesd · 2009-09-07 11:15:00

mangus wrote:

oztrailrider wrote:
Allan wrote:
Hmmm.... I am interested to see the outcome of this: http://lwn.net/Articles/351058/
CK's replied.. http://thread.gmane.org/gmane.linux.kernel/886319
Hero!! LOL! The fight is Back , too...

In Con's response:

Con wrote:

Do you know what a normal desktop PC looks like? No, a more realistic
question based on what you chose to benchmark to prove your point
would be: Do you know what normal people actually do on them?

Con makes a good point. He specifically mentioned he was targeting desktop performance. How many people run a "dual quad core system with hyperthreading" as a desktop?

It really is a shame two dedicated hackers such as Ingo and Con can't find common ground on this.

samwise · 2009-09-07 12:57:02

I think the reason behind much of the support Con has is that he is such an 'everyman' figure - he's not even a professional programmer. If he can do it, there's hope all of us might one day be hero kernel hackers. Good on him. Even if none of his code made the mainline, he has made a difference.

But it's always a shame to see politics - and it is politics - in the kernel development progress. The Linux development model is the worst kind of development model, except for all the other development models we've tried .

I'm going to build a patched kernel and check this out, just to remind myself that the mainline kernel is not the be-all and end-all of Linux. I'm skeptical there will be any objective improvements, but he seems to have made the scheduling code a lot smaller, and simple is good.

andre.ramaciotti · 2009-09-07 13:33:54

I don't get Con's reaction to Ingo's message. Con made a new scheduler for the Linux kernel, so it's frigging obvious that someone would say something about it in the mailing list. If it weren't Ingo, then it would be someone else.

I know they had some conflicts before, but the way Con answered Ingo's benchmarks was plain stupid. He could had answered simply 'Hi Ingo, this isn't the kind of computer I had in mind when I developed BFS and here are the benchmarks I did that I find more useful'. Is it this hard? This way there would be a discussion about the results and not about who's being the troll in this case.

That being said, I'm off to compile a custom kernel.

cerbie · 2009-09-07 16:31:31

I was saddened when the .30 kernels finally got power management working perfectly, but the rt kernel wasn't updated (I've tried random daily versions, before, and had problems, so...). But, almost any page file accesses seem to turn my old P3 notebook to mush (with a non-RT kernel), and any read and writing at the same time dogs it down a bit (such as extracting archives). Turning all swap off guarantees that any web browser will crash on me, so is not a good option. Between that and no Vsync, I've been using Windows much more, lately, since it has Vsync and doesn't get quite as doggy. But, between normal Windows behavior, and idling using over half my RAM, it's always somewhat pokey. Aarrrgh!

Time to make me up a new patched kernel, and see what happens!

As far as the ego war...*sigh* it basically is just that. Fun to read, but honestly, both of them could stand to sit down and start by defining use situations (thus benchmarks to care about) and hardware configurations to care about (16-core monster != normal desktop, TCP latency is always good enough, etc.).

bangkok_manouel · 2009-09-07 16:56:41

2.6.31-rc9-sched-bfs-210.patch is out!

.:B:. · 2009-09-07 17:13:41

Let us know how it goes . For the moment I'll be watching from the sideline.

* B is a chicken

bangkok_manouel · 2009-09-07 17:15:50

building gcc at this moment to see if it'll survive....

edit: ==> Finished making: gcc 4.4.1-1 i686 (Tue Sep 8 00:52:54 ICT 2009)

Last edited by bangkok_manouel (2009-09-07 17:53:28)

makimaki · 2009-09-07 17:23:33

bangkok_manouel wrote:

2.6.31-rc9-sched-bfs-210.patch is out!

Yay!

that's what I've been waiting for, I'm currently using kernel26-git from AUR as I get better performance for my intel 945gme on .31.

Will compile this eve on my netbook and see how it runs

Arch Linux

#26 2009-09-07 03:15:59

Re: omg! Con Kolivas is back!

#27 2009-09-07 04:19:23

Re: omg! Con Kolivas is back!

#28 2009-09-07 04:22:30

Re: omg! Con Kolivas is back!

#29 2009-09-07 04:40:01

Re: omg! Con Kolivas is back!

#30 2009-09-07 06:13:13

Re: omg! Con Kolivas is back!

#31 2009-09-07 06:16:41

Re: omg! Con Kolivas is back!

#32 2009-09-07 06:29:28

Re: omg! Con Kolivas is back!

#33 2009-09-07 08:27:44

Re: omg! Con Kolivas is back!

#34 2009-09-07 08:58:48

Re: omg! Con Kolivas is back!

#35 2009-09-07 09:16:59

Re: omg! Con Kolivas is back!

#36 2009-09-07 09:25:07

Re: omg! Con Kolivas is back!

#37 2009-09-07 09:31:54

Re: omg! Con Kolivas is back!

#38 2009-09-07 09:52:41

Re: omg! Con Kolivas is back!

#39 2009-09-07 10:13:32

Re: omg! Con Kolivas is back!

#40 2009-09-07 10:16:04

Re: omg! Con Kolivas is back!

#41 2009-09-07 10:22:11

Re: omg! Con Kolivas is back!

#42 2009-09-07 10:40:09

Re: omg! Con Kolivas is back!

#43 2009-09-07 11:15:00

Re: omg! Con Kolivas is back!

#44 2009-09-07 12:57:02

Re: omg! Con Kolivas is back!

#45 2009-09-07 13:33:54

Re: omg! Con Kolivas is back!

#46 2009-09-07 16:31:31

Re: omg! Con Kolivas is back!

#47 2009-09-07 16:56:41

Re: omg! Con Kolivas is back!

#48 2009-09-07 17:13:41

Re: omg! Con Kolivas is back!

#49 2009-09-07 17:15:50

Re: omg! Con Kolivas is back!

#50 2009-09-07 17:23:33

Re: omg! Con Kolivas is back!

Board footer