The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

jakob · 2013-05-09 12:11:32

Hey, thanks for providing repo packages of the new version, @Thaodan!

However, to them, too, the error I reported two posts earlier applies: I can't use any other governor than performance and powersave. Is that on purpose?

Thaodan · 2013-05-10 04:35:54

jakob wrote:

Hey, thanks for providing repo packages of the new version, @Thaodan!
However, to them, too, the error I reported two posts earlier applies: I can't use any other governor than performance and powersave. Is that on purpose?

It isn't I just updated the kernel config so that it's like the arch one except the tuxonice, bfs stuff.

BTW: My Google Docs folder with sandy and ivvybridge version: https://drive.google.com/folderview?id= … sp=sharing
EDIT2: Do you use an intel sandybridge cpu?

Last edited by Thaodan (2013-05-10 05:39:26)

jakob · 2013-05-10 10:05:27

Thaodan wrote:

It isn't I just updated the kernel config so that it's like the arch one except the tuxonice, bfs stuff.
BTW: My Google Docs folder with sandy and ivvybridge version: https://drive.google.com/folderview?id= … sp=sharing
EDIT2: Do you use an intel sandybridge cpu?

Meanwhile, I built another custom kernel (based on 3.8.2-3 config) where I explicitly took care of the modules and still had to modprobe the modules. Maybe that was what I just missed with your one, too, as I never had to do this before. Will just add them to /etc/modules-load.d.

Yes, I'm using a 2nd generation sandybridge cpu (corei7-avx), however mostly with a core2 optimisation as that was all that was available so far…

Thanks for your reply!

Thaodan · 2013-05-10 12:36:01

Then see this about the cpufreq stuff: http://www.phoronix.com/scan.php?page=n … px=MTI5Mzc

abraxas88 · 2013-05-10 17:29:47

I finally managed to solve my compilation problems, like trizen commented on AUR, you should add bc in the makedependence array. I installed it and everything went cool.

pablox · 2013-05-13 02:53:55

I have an i7-3770k, installing the ivybridge package would give me an advantage over the pf-core2 one?

Last edited by pablox (2013-05-13 02:54:11)

graysky · 2013-05-13 07:21:17

@pablox - You probably want to compile your own enabling optimization for the 3rd gen core processor or the native optimization.

Thaodan · 2013-05-13 20:53:33

graysky wrote:

@pablox - You probably want to compile your own enabling optimization for the 3rd gen core processor or the native optimization.

Or use the packages from my google drive.

Thaodan · 2013-05-14 14:25:02

i try to make a repo at dropbox, But actually I don't get it, to get a shareable link...

Thaodan · 2013-05-14 20:58:25

So the repo is online. The repo includes packages for sandy and ivrybridge intel cpus. Build with stok arch settings with minor version stripped:
http://dl.dropboxusercontent.com/u/1725 … -pf/x86_64

jakob · 2013-05-15 00:27:29

Hey wow, that sounds great! However, the link doesn't work here, I get a 404…!

Thaodan · 2013-05-15 04:37:16

If you need to acces through a browser use this link: http://dl.dropboxusercontent.com/u/1725 … index.html

jakob · 2013-05-17 11:50:00

Alright, so the entry in /etc/pacman.conf should look like follows

[thao]
# Thaodan's linux-pf repository
# x86_64 only, optimized sandybridge and ivybridge kernel packages & nvidia-pf-sandybrigde package
# To see which packages are available, check
# http://dl.dropboxusercontent.com/u/172590784/Linux-pf/x86_64/index.html
SigLevel = Optional TrustedOnly
Server = http://dl.dropboxusercontent.com/u/172590784/Linux-pf/x86_64

Cheers, I'm getting 3.9.2 at the moment and it works great! Thanks!

Last edited by jakob (2013-05-17 12:40:00)

Thaodan · 2013-05-17 12:27:38

yes Thanks.
BTW: Thaodan not Thaodon^^
If someone needs a i686 package I'll build id.

Last edited by Thaodan (2013-05-17 12:30:03)

jakob · 2013-05-17 12:40:28

Alright, corrected. Should I put this information on the linux-pf wiki page and into the list of unofficial repositories?

jakob · 2013-05-18 13:07:17

Wow, double post, but that one's necessary: for everyone who (like me) still hasn't got it what the new intel_pstate driver is all about, check this post. The ondemand governor simply isn't up to date any more, and on sandybridge cpus it is now automatically disabled through the pstate driver. More on that in this Google+ thread. Apparently, performance and powersave are doing a better job…

Thaodan · 2013-05-22 22:26:33

updated, added patch to enanble cpufreq/intelp with ivybridge cpus.
Repo packages are updated too.

demon012 · 2013-05-23 09:41:23

Yesterday I updated to the 3.9 pf kernel and to my horror aufs3 does not seem to be present in it, I was then forced to switch to the aufs3 friendly kernels.

When can we expect to see aufs3 back in the pf kernel?

*EDIT* Ok apparently it is supposed to already be in the kernel as a patch is applied. However in the kernel config I cannot find any mention of AUFS in any menu section or by editing the .config file myself.

*EDIT* Also tried the prebuild pf kernels and they also seem to be lacking AUFS.

Thanks,

demon012

Last edited by demon012 (2013-05-23 10:26:15)

Thaodan · 2013-05-23 13:35:34

Yes it should be included, but I don't why it isn't . All patches are were applied without error.

nous · 2013-05-24 18:05:38

pablox wrote:

I have an i7-3770k, installing the ivybridge package would give me an advantage over the pf-core2 one?

In as few words as it can be stated: NONE.

In more words: NONE, unless you do all day long AVX-accelerated twofish encryption

If someone considers an advantage 1 millisecond (that's 1/1000th of a second) of gain every 5 minutes, then he should use his home PC for bitcoin mining, and I can elaborate on that if it's unclear why.

Thaodan · 2013-05-24 18:13:00

nous wrote:

pablox wrote:
I have an i7-3770k, installing the ivybridge package would give me an advantage over the pf-core2 one?
In as few words as it can be stated: NONE.
In more words: NONE, unless you do all day long AVX-accelerated twofish encryption
If someone considers an advantage 1 millisecond (that's 1/1000th of a second) of gain every 5 minutes, then he should use his home PC for bitcoin mining, and I can elaborate on that if it's unclear why.

Then you should stop building any build that is not the generic one, if use an optimized build use that build that fits best to your CPU.

nous · 2013-05-24 20:52:04

Thaodan wrote:

nous wrote:
pablox wrote:
I have an i7-3770k, installing the ivybridge package would give me an advantage over the pf-core2 one?
In as few words as it can be stated: NONE.
In more words: NONE, unless you do all day long AVX-accelerated twofish encryption
If someone considers an advantage 1 millisecond (that's 1/1000th of a second) of gain every 5 minutes, then he should use his home PC for bitcoin mining, and I can elaborate on that if it's unclear why.
Then you should stop building any build that is not the generic one, if use an optimized build use that build that fits best to your CPU.

This has been discussed time and again... The kernel developers know better than any of us when it's worth adding a new CPU family into the kernel config, as this was the case with the existing options. I've seen benchmarks (I think graysky posted some recently) comparing generic kernels vs AVX ones and the speed gain was insignificant. It's completely safe to assume that a Core 2 vs AVX benchmark would show an even smaller margin (if any).

Therefore, yes, I'm building kernels optimized for CPUs that are included upstream because I trust people with better knowledge than mine (and only as a service to the repo users). And of course, if someone else compiles a kernel optimized specifically for my CPU I will and shall use it, but personally I wouldn't spend half an hour compiling an AVX kernel for my AVX box to only gain 1 second in total until I'll have to compile the next updated linux-pf.

Last edited by nous (2013-05-24 21:08:23)

Thaodan · 2013-05-24 23:22:07

nous wrote:

Thaodan wrote:
nous wrote:
In as few words as it can be stated: NONE.
In more words: NONE, unless you do all day long AVX-accelerated twofish encryption
If someone considers an advantage 1 millisecond (that's 1/1000th of a second) of gain every 5 minutes, then he should use his home PC for bitcoin mining, and I can elaborate on that if it's unclear why.
Then you should stop building any build that is not the generic one, if use an optimized build use that build that fits best to your CPU.
This has been discussed time and again... The kernel developers know better than any of us when it's worth adding a new CPU family into the kernel config, as this was the case with the existing options. I've seen benchmarks (I think graysky posted some recently) comparing generic kernels vs AVX ones and the speed gain was insignificant. It's completely safe to assume that a Core 2 vs AVX benchmark would show an even smaller margin (if any).
Therefore, yes, I'm building kernels optimized for CPUs that are included upstream because I trust people with better knowledge than mine (and only as a service to the repo users). And of course, if someone else compiles a kernel optimized specifically for my CPU I will and shall use it, but personally I wouldn't spend half an hour compiling an AVX kernel for my AVX box to only gain 1 second in total until I'll have to compile the next updated linux-pf.

If you say cause the kernel developers now it better than you should only use the vanilla kernel, cause the now it better...

nous · 2013-05-25 10:17:51

Thaodan wrote:

If you say cause the kernel developers now it better than you should only use the vanilla kernel, cause the now it better...

Please, stop acting like a troll and responding with "you should" one-liners.

The BFS kernels are by definition a little slower than vanilla, which is the trade-off for the lower latency (i.e. better responsiveness) they provide by default. That been said, it's also possible to tweak the CFS scheduler towards lower latencies but my subjective perception judged it poorly compared to BFS.

graysky · 2013-05-25 10:31:13

nous wrote:

The BFS kernels are by definition a little slower than vanilla, which is the trade-off for the lower latency (i.e. better responsiveness) they provide by default. That been said, it's also possible to tweak the CFS scheduler towards lower latencies but my subjective perception judged it poorly compared to BFS.

Have to disagree with you here, nous. Three endpoints including time to compress, time to compile, and time to encode (x264) where evaluated and BFS patched kernels were faster on the whole than their non-bfs patched counterparts.

http://repo-ck.com/bench/cpu_schedulers_compared.pdf

"In addition to the primary design goals of the bfs, increased desktop interactivity and responsiveness, kernels patched with the ck1 patch set including the bfs outperformed the vanilla kernel using the cfs at nearly all the performance-based benchmarks tested. Further study with a larger test set could be conducted, but based on the small test set of 7 PCs evaluated, these increases in process queuing, efficiency/speed are, on the whole, independent of CPU type (mono, dual, quad, hyperthreaded, etc.), CPU architecture (32-bit and 64-bit), and of CPU multiplicity (mono or dual socket)."

Last edited by graysky (2013-05-25 10:32:53)

Arch Linux

#701 2013-05-09 12:11:32

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#702 2013-05-10 04:35:54

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#703 2013-05-10 10:05:27

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#704 2013-05-10 12:36:01

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#705 2013-05-10 17:29:47

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#706 2013-05-13 02:53:55

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#707 2013-05-13 07:21:17

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#708 2013-05-13 20:53:33

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#709 2013-05-14 14:25:02

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#710 2013-05-14 20:58:25

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#711 2013-05-15 00:27:29

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#712 2013-05-15 04:37:16

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#713 2013-05-17 11:50:00

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#714 2013-05-17 12:27:38

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#715 2013-05-17 12:40:28

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#716 2013-05-18 13:07:17

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#717 2013-05-22 22:26:33

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#718 2013-05-23 09:41:23

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#719 2013-05-23 13:35:34

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#720 2013-05-24 18:05:38

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#721 2013-05-24 18:13:00

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#722 2013-05-24 20:52:04

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#723 2013-05-24 23:22:07

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#724 2013-05-25 10:17:51

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

#725 2013-05-25 10:31:13

Re: The linux-pf thread; BFS/CK, TuxOnIce, BFQ, AUFS3

Board footer