Kernel26-ck with BFS

smakked · 2010-03-08 04:22:27

Is anyone using this kernel?
I messed around with it over the weekend and settled on my MHZ set at 300 and i am impressed with its responsivness.
Makes KDE4 alot smoother and snappier. Anyone else use it and what have you found.?

Last edited by smakked (2010-04-15 08:56:13)

bangkok_manouel · 2010-03-08 07:00:34

http://www.google.com/search?q=ck+bfs+s … hlinux.org

Anarconda · 2010-03-08 09:29:21

The only problem I found is I can compile/load 'vboxdrv' without errors but VirtualBox can't find it.

matkam · 2010-04-15 03:59:10

I just installed it and it seems good. Compiling a large package with -j8 makeflag puts all 8 of my logical CPUs to 100% utilization instead of around 98%. And system still feels usable and snappy.

Anarconda wrote:

The only problem I found is I can compile/load 'vboxdrv' without errors but VirtualBox can't find it.

I hear Virtualbox has an issue with higher timing frequencies. http://bbs.archlinux.org/viewtopic.php? … 22#p704522

Last edited by matkam (2010-04-15 04:50:58)

fabertawe · 2010-04-15 08:50:12

I compiled it with "CONFIG_HZ=1000" and for my CPU and it feels ever so slightly snappier than the stock kernel but certainly more responsive under full load. No problems compiling the modules for VirtualBox or running it.

broch · 2010-04-15 12:48:19

obviously, unless you remove dynticks it does not matter what HZ value is set

SanskritFritz · 2010-04-15 14:42:45

broch wrote:

obviously, unless you remove dynticks it does not matter what HZ value is set

Are you sure? Dynticks only defines, that the timer frequency can be lowered, but a maximum speed can be set nevertheless.
http://lwn.net/Articles/223185/

An idle CPU can save quite a bit of power, but waking that CPU up 100 times (or more) per second will hurt those power savings considerably. With a flexible timer infrastructure, there is no point in turning the CPU back on until it has something to do. So, when the (i386) kernel goes into its idle loop, it checks the next pending timer event. If that event is further away than the next tick, the periodic tick is turned off altogether; instead, the timer is is programmed to fire when the next event comes due. The CPU can then rest unharrassed until that time - unless an interrupt comes in first. Once the processor goes out of the idle state, the periodic tick is restored.

broch · 2010-04-15 15:29:04

Yes, now read this again:
How can you set CONFIG_HZ=3000 and dynamic ticks ath the same time? Dynticks will set dynamic interrupts (ondemand) CONFIG_HZ=3000 will force defined frequency of interrupts. These two are mutually exclusive.

If you have laptop it is easy to test the consequences
set CONFIG_HZ=2000
disable dynticks
boot kernel with above settings and observe cpu temp; It will go up to the roof because CPU will be constantly working whether this is needed or not,
Now set CONFIG_HZ=2000 with enabled dynticks: CPU temp will stay at low.

This is the most obvious consequence of playing with CONFIG_HZ and dynticks. In other words, with dynticks enabled you can safely set CONFIG_HZ=10000 and nothing will happen. So if someone is claiming otherwise - that changing CONFIG_HZ with enabled dynticks did something to OS responsiveness is a victim of placebo effect. Of course placebo is not bad if making user experience better (even if not real)

bzipitidoo · 2010-04-15 17:03:55

Dynamic ticks and tick frequency are not necessary exclusive. From what I've heard, choosing a particular tick frequency causes other optimizations to be chosen. Even if dynamic ticks is enabled, choosing a low HZ will select optimizations suitable for a server, and a high HZ will select optimizations suitable for a desktop. Users could certainly notice a difference.

broch · 2010-04-15 23:13:49

Just try it. For server low values are better so no need for dynticks. Good desktop with proper cooling dopnkt really needs dynticks. These are mostly useful for laptops (battery and poor cooling), but desktop will do well also.
You can't have both: one will take over. It is more or less like being programmed for walk and stop. Possible? Yes but you can' t do both at the same time. In this case dynticks are safer so they take over irrelevant of CONFIG_HZ set.

SanskritFritz · 2010-04-16 06:52:26

broch wrote:

Yes, now read this again:
How can you set CONFIG_HZ=3000 and dynamic ticks ath the same time? Dynticks will set dynamic interrupts (ondemand) CONFIG_HZ=3000 will force defined frequency of interrupts. These two are mutually exclusive.
...
This is the most obvious consequence of playing with CONFIG_HZ and dynticks. In other words, with dynticks enabled you can safely set CONFIG_HZ=10000 and nothing will happen.

That just means, that with dynamic ticks turned off the CPU will not go to idle state, while when turned on it does. Setting the frequency merely sets the maximum tick frequency ti use when the CPU is under load.

broch · 2010-04-16 10:02:40

I would suggest testing. You then you will see a difference.
I hope that you understand that citing paper from 2007 when dynticks were not yet fully implemented has not much value?

beroal · 2010-04-16 11:33:22

broch wrote:

I would suggest testing. You then you will see a difference.

Also inspecting the kernel code may help. (Sorry for interruption. I'm considering real-time kernels to stabilize media players work, but can not decide, should I use "rt" or "ck" or something else.)

mechmg93 · 2010-04-16 11:55:13

i ve been using kernel26-ck for a few hours and i noticed that is very swap aggressive. I did not mentioned any great difference from stock kernel in multitasking usage until now.

the kernel is installed in my lenovo r61 laptop.

Last edited by mechmg93 (2010-04-16 11:55:43)

SanskritFritz · 2010-04-17 09:57:16

I recompiled the kernel26-ck with Timer frequency 1000 HZ instead of the default 300 HZ and have not noticed any improvement at all. I admit, I left Dynamic Ticks on ;-)

falconindy · 2010-04-17 18:45:57

Hmm... this might explain why VBox starting throwing kernel panics at me when I compiled my kernel with 10000Hz and decided to disable dyn ticks. ck's documentation (albeit dated) claims that you should be disabling it.

SanskritFritz · 2010-04-17 19:19:22

falconindy wrote:

ck's documentation (albeit dated) claims that you should be disabling it.

http://ck.kolivas.org/patches/bfs/bfs-faq.txt wrote:

THESE ARE OPTIONAL FOR LOWEST LATENCY. YOU DO NOT NEED THESE!
Configure your kernel with 1000Hz, preempt ON and disable dynamic ticks.

falconindy · 2010-04-17 21:51:58

Fritz, your avatar never fails to give me to creeps.

I'm not as well educated on the inner workings on the kernel as I'd like to be. Any idea why this might be recommend this over a higher clock frequency with dynamic ticks? Or rather, what would be the potential benefit to a higher clock frequency, under the assumption that your machine remains stable?

pogeymanz · 2010-04-27 13:53:15

This is directly from the kernel config help menu for 1000Hz:

1000 Hz is the preferred choice for desktop systems and other systems requiring fast interactive responses to events. 
Laptops can also benefit from this choice without sacrificing battery life if dynticks is also enabled.

So, it would appear that setting a frequency and dynamic ticks are not mutually exclusive.

I'm going to see if 2000Hz makes any interactivity difference while under load. Of course this could be placebo, but I'll try to be as objective as I know how.

One this is for sure, BFS does make a huge difference when my CPU is cranking away at compiling or running Matlab scripts.

Update: It seems like 2000Hz really does make a difference. My major test for this kind of stuff is to compile the kernel and while it's compiling, I scroll through my Firefox tabs, which usually stutters, launch dmenu-launch, and browse through pictures, etc in Thunar. I have gotten NOTHING that would indicate my CPU is under load. Now I want to time the compile to see if there is a big difference in throughput. There must be, right?

Last edited by pogeymanz (2010-04-27 14:21:18)

eduardo.eae · 2010-04-27 14:16:51

I use a frequency of 100Hz, with dynticks on, and it reduced to half (50) the wakeups-from-idle(powertop) on my laptop. And my system is very responsive, even under high load.

broch · 2010-04-27 16:31:37

pogeymanz wrote:

This is directly from the kernel config help menu for 1000Hz:
1000 Hz is the preferred choice for desktop systems and other systems requiring fast interactive responses to events. 
Laptops can also benefit from this choice without sacrificing battery life if dynticks is also enabled.
So, it would appear that setting a frequency and dynamic ticks are not mutually exclusive.
I'm going to see if 2000Hz makes any interactivity difference while under load. Of course this could be placebo, but I'll try to be as objective as I know how.
One this is for sure, BFS does make a huge difference when my CPU is cranking away at compiling or running Matlab scripts.
Update: It seems like 2000Hz really does make a difference. My major test for this kind of stuff is to compile the kernel and while it's compiling, I scroll through my Firefox tabs, which usually stutters, launch dmenu-launch, and browse through pictures, etc in Thunar. I have gotten NOTHING that would indicate my CPU is under load. Now I want to time the compile to see if there is a big difference in throughput. There must be, right?

SanskritFritz cited:

http://ck.kolivas.org/patches/bfs/bfs-faq.txt wrote:
THESE ARE OPTIONAL FOR LOWEST LATENCY. YOU DO NOT NEED THESE!
Configure your kernel with 1000Hz, preempt ON and disable dynamic ticks.

out of curiosity: are you discussing Kolivas guidelines? He is responsible for BFS.
Why don't you read Kolivas FAQ if you are using his scheduler?
In general higher CONFIG_HZ (without dynticks) more unstable system (crashing apps, WM/DM and so on).

dynticks and CONFIG_HZ are not exclusive: if dynticks are set this is a mechanism that governs interrupts HZ.
If you want real test compile something demanding: big stuff
heh, you could try firefox icc with -ipo1

Last edited by broch (2010-04-27 16:33:41)

gtklocker · 2010-04-27 18:25:00

What is good for a laptop? (low-rage)

SanskritFritz · 2010-04-27 19:18:29

gtklocker wrote:

What is good for a laptop? (low-rage)

Dynamic ticks are definitely recommended, and possibly low frequency for saving the battery.

flamelab · 2010-04-27 19:42:05

gtklocker wrote:

What is good for a laptop? (low-rage)

By disabling dynticks, CPU is always "awake". So that's not good for your system consumption. But it may increase system responsiveness.

gtklocker · 2010-04-27 19:50:07

Hm, thanks. What CONFIG_HZ should I set?

processor       : 1
vendor_id       : AuthenticAMD
cpu family      : 17
model           : 3
model name      : AMD Turion(tm)X2 Dual Core Mobile RM-70
stepping        : 1
cpu MHz         : 500.000
cache size      : 512 KB
physical id     : 0
siblings        : 2
core id         : 1
cpu cores       : 2
apicid          : 1
initial apicid  : 1
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt rdtscp lm 3dnowext 3dnow constant_tsc rep_good nonstop_tsc extd_apicid pni cx16 lahf_lm cmp_legacy svm extapic cr8_legacy 3dnowprefetch osvw skinit lbrv svm_lock nrip_save
bogomips        : 3991.46
TLB size        : 1024 4K pages
clflush size    : 64
cache_alignment : 64
address sizes   : 40 bits physical, 48 bits virtual
power management: ts ttp tm stc 100mhzsteps hwpstate

Last edited by gtklocker (2010-04-27 19:50:27)

Arch Linux

#1 2010-03-08 04:22:27

Kernel26-ck with BFS

#2 2010-03-08 07:00:34

Re: Kernel26-ck with BFS

#3 2010-03-08 09:29:21

Re: Kernel26-ck with BFS

#4 2010-04-15 03:59:10

Re: Kernel26-ck with BFS

#5 2010-04-15 08:50:12

Re: Kernel26-ck with BFS

#6 2010-04-15 12:48:19

Re: Kernel26-ck with BFS

#7 2010-04-15 14:42:45

Re: Kernel26-ck with BFS

#8 2010-04-15 15:29:04

Re: Kernel26-ck with BFS

#9 2010-04-15 17:03:55

Re: Kernel26-ck with BFS

#10 2010-04-15 23:13:49

Re: Kernel26-ck with BFS

#11 2010-04-16 06:52:26

Re: Kernel26-ck with BFS

#12 2010-04-16 10:02:40

Re: Kernel26-ck with BFS

#13 2010-04-16 11:33:22

Re: Kernel26-ck with BFS

#14 2010-04-16 11:55:13

Re: Kernel26-ck with BFS

#15 2010-04-17 09:57:16

Re: Kernel26-ck with BFS

#16 2010-04-17 18:45:57

Re: Kernel26-ck with BFS

#17 2010-04-17 19:19:22

Re: Kernel26-ck with BFS

#18 2010-04-17 21:51:58

Re: Kernel26-ck with BFS

#19 2010-04-27 13:53:15

Re: Kernel26-ck with BFS

#20 2010-04-27 14:16:51

Re: Kernel26-ck with BFS

#21 2010-04-27 16:31:37

Re: Kernel26-ck with BFS

#22 2010-04-27 18:25:00

Re: Kernel26-ck with BFS

#23 2010-04-27 19:18:29

Re: Kernel26-ck with BFS

#24 2010-04-27 19:42:05

Re: Kernel26-ck with BFS

#25 2010-04-27 19:50:07

Re: Kernel26-ck with BFS

Board footer