Official thread: "zen-sources" - Part 7

Message

rahulthewall · Post by **rahulthewall** » Fri Sep 18, 2009 8:52 pm

I am not sure whether anyone noticed this, but Phoronix has run done some benchmarking on BFS.
http://www.phoronix.com/scan.php?page=a ... arks&num=1

Jupiter1TX · Post by **Jupiter1TX** » Fri Sep 18, 2009 11:17 pm

I ran some of the same test as in the Phoronix test.
As you can see, results are mixed. I also named the
test incorrectly, it should be zen1 not zen0. Once i am
able to get logged in to PST i will try and update this
and other test.

Code: Select all

====================================
Timed PHP Compilation 5.2.9
Time To Compile
====================================

Linux-2.6.31-zen0+CFS: 25.73
Linux-2.6.31-zen0+BFS: 25.65


====================================
Timed Apache Compilation 2.2.11
Time To Compile
====================================

Linux-2.6.31-zen0+CFS: 3.48
Linux-2.6.31-zen0+BFS: 3.41


====================================
7-Zip Compression 4.65
Compress Speed Test
====================================

Linux-2.6.31-zen0+CFS: 18404.00
Linux-2.6.31-zen0+BFS: 17836.66


====================================
Apache Benchmark 2.2.11
Static Web Page Serving
====================================

Linux-2.6.31-zen0+CFS: 8656.34
Linux-2.6.31-zen0+BFS: 13692.08


=====================================================================
Threaded I/O Tester 0.3.3
Test Item: Random Write - Filesize Per Thread: 16MB - Thread Count: 8
=====================================================================

Linux-2.6.31-zen0+CFS: 3.00
Linux-2.6.31-zen0+BFS: 4.00


=============================================================
Threaded I/O Tester 0.3.3
Test Item: Read - Filesize Per Thread: 16MB - Thread Count: 8
=============================================================

Linux-2.6.31-zen0+CFS: 2.00
Linux-2.6.31-zen0+BFS: 2.00


====================================
PostMark 1.51
Disk Transaction Performance
====================================

Linux-2.6.31-zen0+CFS: 1063.50
Linux-2.6.31-zen0+BFS: 967.25

HecHacker1 · Post by **HecHacker1** » Sat Sep 19, 2009 12:23 am

Regarding the Phoronix BFS vs mainline benchmarks, lots of discussion is happening on the linux kernel mailing lists.

The gist of it is that CFS can perform better than BFS, but only with changes to the many scheduler sys controls in /proc/sys/kernel/.

If you want to go through the whole thread:

http://lkml.org/lkml/2009/9/6/136

You'll see lots of discussion, with many patches to CFS after discovering BFS did much better at some tasks.

That's good news, and we probably would have never seen these patches without BFS as competition.

Overall it seems BFS isn't particularly bad at anything, and yet is a much simpler scheduler with fewer tuneables.

Maybe we can get the Zen dev's to pull the latest CFS changes? And perhaps use some of the suggested tuneables in the threads.

I'll be sticking with BFS though just to test it out. Con is still improving it, he's on BFS v222-test4 right now, which seems to further simplify and fix fairness issues.

pappy_mcfae · Post by **pappy_mcfae** » Sat Sep 19, 2009 5:00 am

heavyjoost wrote:
pappy_mcfae wrote:No kidding. This machine loves BFS. My future web server loves bfs. My multi-processor machines don't...whether x86 or x86_64. I keep waiting in hope that one of the heads-up emails will find me staring at a patch that will make BFS work on my multiprocessor machines.

Blessed be!
Pappy
No problems here and I'm running on an dual-core AMD CPU. Running on Debian (testing) though.

Please post the .config you have for this. I'd be interested to see it.

Blessed be!
Pappy

ponciarello · Post by **ponciarello** » Sat Sep 19, 2009 5:34 am

if it can be useful for you, this is a .config for toshiba satellite X200-24E (dual core intel) and this is of an amd assembled pc.

rahulthewall · Post by **rahulthewall** » Sat Sep 19, 2009 8:11 am

And here is the config for my Dell Latitude D520 (old laptop with Intel T2300). In case it helps you.

Ant P. · Post by **Ant P.** » Sat Sep 19, 2009 11:15 am

My wifi crashes are back again. Using this box as a hostapd access point, and it dies if I try to connect to it.

Code: Select all

Sep 19 01:05:54 [kernel] [  170.968470] ------------[ cut here ]------------
Sep 19 01:05:54 [kernel] [  170.968479] WARNING: at net/mac80211/tx.c:1325 ieee80211_tx+0x23f/0x2d9()
Sep 19 01:05:54 [kernel] [  170.968485] Hardware name:  
Sep 19 01:05:54 [kernel] [  170.968493] Modules linked in: ipt_REJECT xt_tcpudp iptable_filter ip_tables ip6table_filter ip6_tables x_tables sr_mod p54usb p54common cdrom [last unloaded: microcode]
Sep 19 01:05:54 [kernel] [  170.968522] Pid: 0, comm: swapper Tainted: G        W  2.6.31-zen0 #17
Sep 19 01:05:54 [kernel] [  170.968527] Call Trace:
Sep 19 01:05:54 [kernel] [  170.968531]  <IRQ>  [<ffffffff8103486e>] warn_slowpath_common+0x88/0xb6
Sep 19 01:05:54 [kernel] [  170.968547]  [<ffffffff8103491f>] warn_slowpath_fmt+0x4b/0x61
Sep 19 01:05:54 [kernel] [  170.968558]  [<ffffffff813292ee>] ieee80211_tx+0x23f/0x2d9
Sep 19 01:05:54 [kernel] [  170.968570]  [<ffffffff8132954f>] ieee80211_tx_pending+0x1c7/0x2de
Sep 19 01:05:54 [kernel] [  170.968580]  [<ffffffff8133ade2>] ? _spin_unlock+0x3d/0x5c
Sep 19 01:05:54 [kernel] [  170.968590]  [<ffffffff810397d2>] tasklet_action+0x7e/0xe0
Sep 19 01:05:54 [kernel] [  170.968600]  [<ffffffff8103a0e6>] __do_softirq+0xa5/0x151
Sep 19 01:05:54 [kernel] [  170.968609]  [<ffffffff8101f5aa>] ? apic_write+0x24/0x3a
Sep 19 01:05:54 [kernel] [  170.968618]  [<ffffffff8100ce2c>] call_softirq+0x1c/0x28
Sep 19 01:05:54 [kernel] [  170.968628]  [<ffffffff8100ee9f>] do_softirq+0x43/0x98
Sep 19 01:05:54 [kernel] [  170.968637]  [<ffffffff8103a35b>] irq_exit+0x52/0xb6
Sep 19 01:05:54 [kernel] [  170.968646]  [<ffffffff8100e55a>] do_IRQ+0xb3/0xe0
Sep 19 01:05:54 [kernel] [  170.968655]  [<ffffffff8100c693>] ret_from_intr+0x0/0xa
Sep 19 01:05:54 [kernel] [  170.968661]  <EOI>  [<ffffffff81013b45>] ? mwait_idle+0x76/0x8f
Sep 19 01:05:54 [kernel] [  170.968676]  [<ffffffff81013b38>] ? mwait_idle+0x69/0x8f
Sep 19 01:05:54 [kernel] [  170.968685]  [<ffffffff8100b06a>] ? cpu_idle+0xaa/0x10e
Sep 19 01:05:54 [kernel] [  170.968694]  [<ffffffff8132e64e>] ? rest_init+0x92/0xa8
Sep 19 01:05:54 [kernel] [  170.968704]  [<ffffffff814c5a2b>] ? start_kernel+0x317/0x338
Sep 19 01:05:54 [kernel] [  170.968714]  [<ffffffff814c4e61>] ? x86_64_start_reservations+0x91/0xac
Sep 19 01:05:54 [kernel] [  170.968725]  [<ffffffff814c4f6d>] ? x86_64_start_kernel+0xf1/0x10c
Sep 19 01:05:54 [kernel] [  170.968732] ---[ end trace 0b83a2ce3dfaf3d9 ]---

Switching schedulers made no difference here.

I noticed the line "using mwait in idle threads." in my boot messages, does that mean there's possibly a way to turn it off and avoid this crash?

edit: Answered my own question - yes there is (idle=halt) and yes it fixes it.
edit 2: I was wrong... that only delayed it. I've recompiled and turned off "enable wireless powersave by default" in the networking options to see if that was it.
edit 3: Nope. I give up, going back to .31 vanilla for now though I'm not getting my hopes up for that either...

cheater1034 · Post by **cheater1034** » Sat Sep 19, 2009 4:44 pm

HecHacker1 wrote:Regarding the Phoronix BFS vs mainline benchmarks, lots of discussion is happening on the linux kernel mailing lists.

The gist of it is that CFS can perform better than BFS, but only with changes to the many scheduler sys controls in /proc/sys/kernel/.

If you want to go through the whole thread:

http://lkml.org/lkml/2009/9/6/136

You'll see lots of discussion, with many patches to CFS after discovering BFS did much better at some tasks.

That's good news, and we probably would have never seen these patches without BFS as competition.

Overall it seems BFS isn't particularly bad at anything, and yet is a much simpler scheduler with fewer tuneables.

Maybe we can get the Zen dev's to pull the latest CFS changes? And perhaps use some of the suggested tuneables in the threads.

I'll be sticking with BFS though just to test it out. Con is still improving it, he's on BFS v222-test4 right now, which seems to further simplify and fix fairness issues.

Zen has the latest cfs (almost, there are new updates to add now)

The consensus i have been exposed to on irc and have read is that every test shows cfs failing miserably in practical desktop situations.

However, current CFS shows minor improvement, and disabling the features LB_BIAS and NEXT_BUDDY have yielded some positive results - but when disabling features like this it should be considered that these features were added to fix something else, but ended up breaking another thing. (in zen now there is the boost nice option along with latest sched-core-for-linus, and the LB_BIAS and NEXT_BUDDY features have been disabled) - but the fact of the matter is that CFS has not followed a design path for the common desktop system, and bfs clearly exposes this.

Now, bfs is still just for testing and hasn't technically been released - and it can't yet be considered stable - so it still doesn't work for everyone :O

cheater1034 · Post by **cheater1034** » Sat Sep 19, 2009 4:45 pm

Ant_P wrote:My wifi crashes are back again. Using this box as a hostapd access point, and it dies if I try to connect to it.

Code: Select all

Sep 19 01:05:54 [kernel] [  170.968470] ------------[ cut here ]------------
Sep 19 01:05:54 [kernel] [  170.968479] WARNING: at net/mac80211/tx.c:1325 ieee80211_tx+0x23f/0x2d9()
Sep 19 01:05:54 [kernel] [  170.968485] Hardware name:  
Sep 19 01:05:54 [kernel] [  170.968493] Modules linked in: ipt_REJECT xt_tcpudp iptable_filter ip_tables ip6table_filter ip6_tables x_tables sr_mod p54usb p54common cdrom [last unloaded: microcode]
Sep 19 01:05:54 [kernel] [  170.968522] Pid: 0, comm: swapper Tainted: G        W  2.6.31-zen0 #17
Sep 19 01:05:54 [kernel] [  170.968527] Call Trace:
Sep 19 01:05:54 [kernel] [  170.968531]  <IRQ>  [<ffffffff8103486e>] warn_slowpath_common+0x88/0xb6
Sep 19 01:05:54 [kernel] [  170.968547]  [<ffffffff8103491f>] warn_slowpath_fmt+0x4b/0x61
Sep 19 01:05:54 [kernel] [  170.968558]  [<ffffffff813292ee>] ieee80211_tx+0x23f/0x2d9
Sep 19 01:05:54 [kernel] [  170.968570]  [<ffffffff8132954f>] ieee80211_tx_pending+0x1c7/0x2de
Sep 19 01:05:54 [kernel] [  170.968580]  [<ffffffff8133ade2>] ? _spin_unlock+0x3d/0x5c
Sep 19 01:05:54 [kernel] [  170.968590]  [<ffffffff810397d2>] tasklet_action+0x7e/0xe0
Sep 19 01:05:54 [kernel] [  170.968600]  [<ffffffff8103a0e6>] __do_softirq+0xa5/0x151
Sep 19 01:05:54 [kernel] [  170.968609]  [<ffffffff8101f5aa>] ? apic_write+0x24/0x3a
Sep 19 01:05:54 [kernel] [  170.968618]  [<ffffffff8100ce2c>] call_softirq+0x1c/0x28
Sep 19 01:05:54 [kernel] [  170.968628]  [<ffffffff8100ee9f>] do_softirq+0x43/0x98
Sep 19 01:05:54 [kernel] [  170.968637]  [<ffffffff8103a35b>] irq_exit+0x52/0xb6
Sep 19 01:05:54 [kernel] [  170.968646]  [<ffffffff8100e55a>] do_IRQ+0xb3/0xe0
Sep 19 01:05:54 [kernel] [  170.968655]  [<ffffffff8100c693>] ret_from_intr+0x0/0xa
Sep 19 01:05:54 [kernel] [  170.968661]  <EOI>  [<ffffffff81013b45>] ? mwait_idle+0x76/0x8f
Sep 19 01:05:54 [kernel] [  170.968676]  [<ffffffff81013b38>] ? mwait_idle+0x69/0x8f
Sep 19 01:05:54 [kernel] [  170.968685]  [<ffffffff8100b06a>] ? cpu_idle+0xaa/0x10e
Sep 19 01:05:54 [kernel] [  170.968694]  [<ffffffff8132e64e>] ? rest_init+0x92/0xa8
Sep 19 01:05:54 [kernel] [  170.968704]  [<ffffffff814c5a2b>] ? start_kernel+0x317/0x338
Sep 19 01:05:54 [kernel] [  170.968714]  [<ffffffff814c4e61>] ? x86_64_start_reservations+0x91/0xac
Sep 19 01:05:54 [kernel] [  170.968725]  [<ffffffff814c4f6d>] ? x86_64_start_kernel+0xf1/0x10c
Sep 19 01:05:54 [kernel] [  170.968732] ---[ end trace 0b83a2ce3dfaf3d9 ]---

Switching schedulers made no difference here.

I noticed the line "using mwait in idle threads." in my boot messages, does that mean there's possibly a way to turn it off and avoid this crash?

edit: Answered my own question - yes there is (idle=halt) and yes it fixes it.
edit 2: I was wrong... that only delayed it. I've recompiled and turned off "enable wireless powersave by default" in the networking options to see if that was it.
edit 3: Nope. I give up, going back to .31 vanilla for now though I'm not getting my hopes up for that either...

You need to go to #zen-sources on irc.freenode.org and talk to dodo1122 or anyone else and you can get sorted out quicker

Jupiter1TX · Post by **Jupiter1TX** » Sat Sep 19, 2009 6:26 pm

Not sure why but two times now while using 2.6.31-zen1+BFS
my /usr/portage index gets corrupted when trying to update
using eix-sync and, i have had to recreate my portage partition.
This has never happened without BFS. Am using reiserfs3 and
Funtoo.

aTan · Post by **aTan** » Sat Sep 19, 2009 6:49 pm

Jupiter1TX wrote:Not sure why but two times now while using 2.6.31-zen1+BFS
my /usr/portage index gets corrupted when trying to update
using eix-sync and, i have had to recreate my portage partition.
This has never happened without BFS. Am using reiserfs3 and
Funtoo.

ReiserFS 3.6 is broken with BFS.

Ant P. · Post by **Ant P.** » Sat Sep 19, 2009 7:07 pm

cheater1034 wrote:You need to go to #zen-sources on irc.freenode.org and talk to dodo1122 or anyone else and you can get sorted out quicker

No need, it seems to be stable again after going back to 2.6.31-zen1 (instead of git HEAD). I don't think I should be bugging people directly about it at the moment, at least until the .32-rc's start coming.

cheater1034 · Post by **cheater1034** » Sat Sep 19, 2009 11:23 pm

Ant_P wrote:
cheater1034 wrote:You need to go to #zen-sources on irc.freenode.org and talk to dodo1122 or anyone else and you can get sorted out quicker
No need, it seems to be stable again after going back to 2.6.31-zen1 (instead of git HEAD). I don't think I should be bugging people directly about it at the moment, at least until the .32-rc's start coming.

/facepalm

if you were using origin/master on git (Default) then no wonder

I NEED you to use origin/master-2.6.31, it's 10000x better than origin/master - considering the fact that it works

Ant P. · Post by **Ant P.** » Sat Sep 19, 2009 11:37 pm

I guess I deserved those crashes for not paying attention.

With 2.6.31-zen1, those warnings are still there and they apparently cause the wifi to stop working. It looks like the crash was something else entirely (maybe triggered by this though), because it's been running for hours with no other problems.

cheater1034 · Post by **cheater1034** » Sun Sep 20, 2009 4:53 am

Ant_P wrote:I guess I deserved those crashes for not paying attention.

With 2.6.31-zen1, those warnings are still there and they apparently cause the wifi to stop working. It looks like the crash was something else entirely (maybe triggered by this though), because it's been running for hours with no other problems.

Please if the bug occurs on master-2.6.31 (which is now labeled as zen2) go into IRC and help find it if it's zen-related !
(trust me, i'm much more concerned with bugs in the stable branches than the dev branches come out) -------- especially 2.6.32-rc, it looks like a handfull

master-2.6.31 just updated with tons more upstream scheduler updates - both masters have CFS boost now too

Rion · Post by **Rion** » Sun Sep 20, 2009 5:10 am

its on master now

Code: Select all

# genkernel --disklabel all
* Gentoo Linux Genkernel; Version 3.4.10.906
* Running with options: --disklabel all

* Linux Kernel 2.6.31-zen0 for x86_64...

why zen0 ?

PS i think i'll move got gentoo-sources if my microphone won't work with zen2.. something is really wrong with alsa in zen

MageSlayer · Post by **MageSlayer** » Sun Sep 20, 2009 6:25 am

FYI.

bfs 220, 221, 230 still make my keyboard hang.
Con and guys on #ck are informed.

aTan · Post by **aTan** » Sun Sep 20, 2009 7:36 am

I guess that thanks to CK they've started to work on better CFS. It's already in Zen.
http://linux.derkeiler.com/Mailing-List ... 06953.html

pappy_mcfae · Post by **pappy_mcfae** » Sun Sep 20, 2009 8:32 am

ponciarello wrote:if it can be useful for you, this is a .config for toshiba satellite X200-24E (dual core intel) and this is of an amd assembled pc.

How did you come up with "# Linux kernel version: 2.6.31-zen2 "Alien Mind Creator"", as in the first .config you posted?

Blessed be!
Pappy

aTan · Post by **aTan** » Sun Sep 20, 2009 8:37 am

pappy_mcfae wrote:
ponciarello wrote:if it can be useful for you, this is a .config for toshiba satellite X200-24E (dual core intel) and this is of an amd assembled pc.
How did you come up with "# Linux kernel version: 2.6.31-zen2 "Alien Mind Creator"", as in the first .config you posted?

Blessed be!
Pappy

checkout master-2.6.31 branch.

razum2um · Post by **razum2um** » Sun Sep 20, 2009 8:39 am

aTan wrote:
Jupiter1TX wrote:Not sure why but two times now while using 2.6.31-zen1+BFS
my /usr/portage index gets corrupted when trying to update
using eix-sync and, i have had to recreate my portage partition.
This has never happened without BFS. Am using reiserfs3 and
Funtoo.
ReiserFS 3.6 is broken with BFS.

Any hope, that it would be fixed?
(I have ReiserFS 3 on /home)

aTan · Post by **aTan** » Sun Sep 20, 2009 9:39 am

razum2um wrote:
aTan wrote:
Jupiter1TX wrote:Not sure why but two times now while using 2.6.31-zen1+BFS
my /usr/portage index gets corrupted when trying to update
using eix-sync and, i have had to recreate my portage partition.
This has never happened without BFS. Am using reiserfs3 and
Funtoo.
ReiserFS 3.6 is broken with BFS.
Any hope, that it would be fixed?
(I have ReiserFS 3 on /home)

Smeg knows. It forced me to use Reiser4. There is only one thing in BFS FAQ about it:

I suggest NOT using this with reiserFS3.6
as it uses a very fragile lockless algorithm, although I have no proof that it
will break.

Ant P. · Post by **Ant P.** » Sun Sep 20, 2009 11:55 am

cheater1034 wrote:Please if the bug occurs on master-2.6.31 (which is now labeled as zen2) go into IRC and help find it if it's zen-related !
(trust me, i'm much more concerned with bugs in the stable branches than the dev branches come out) -------- especially 2.6.32-rc, it looks like a handfull

Hopefully I've got enough free time today, just checked out the right branch this time (not the tagged .31-zen1

). If it gives me problems I'll be there.

ponciarello · Post by **ponciarello** » Sun Sep 20, 2009 12:01 pm

pappy_mcfae wrote:How did you come up with "# Linux kernel version: 2.6.31-zen2 "Alien Mind Creator"", as in the first .config you posted?

issue these commands inside git tree

Code: Select all

git fetch

(for syncing) and

Code: Select all

git diff v2.6.31 origin/master-2.6.31 > ../2.6.31-zen.patch

to make a patch of latest master-2.6.31 (to apply on 2.6.31 vanilla).

cheater1034 · Post by **cheater1034** » Sun Sep 20, 2009 3:02 pm

Rion wrote:its on master now
Code: Select all
# genkernel --disklabel all
* Gentoo Linux Genkernel; Version 3.4.10.906
* Running with options: --disklabel all

* Linux Kernel 2.6.31-zen0 for x86_64...
why zen0 ?

PS i think i'll move got gentoo-sources if my microphone won't work with zen2.. something is really wrong with alsa in zen

http://zen-sources.org/content/master-c ... -using-git

Any why genkernel :O, (the only thing i'd slightly reccomend genkernel for is the initramfs)

This patch should fix the problems with ATI drivers and possibly the i915 drivers too, so you can try it if you are having a problem with either:
http://ck.kolivas.org/patches/bfs/bfs23 ... ers2.patch

Official thread: "zen-sources" - Part 7

BFS Phoronix

Re: BFS Phoronix