Gentoo Forums
Gentoo Forums
Gentoo Forums
Quick Search: in
Love-sources - get them before they go obsolete
View unanswered posts
View posts from last 24 hours

Goto page Previous  1, 2, 3, 4, 5, 6, 7  Next  
Reply to topic    Gentoo Forums Forum Index Off the Wall
View previous topic :: View next topic  
Author Message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Fri Sep 12, 2003 5:14 am    Post subject: Reply with quote

DOH !!!

I brainfarted the patch and got original and modified dir.

I uploaded a new patch now - which should be the right way around.
Back to top
View user's profile Send private message
neenee
Veteran
Veteran


Joined: 20 Jul 2003
Posts: 1786

PostPosted: Fri Sep 12, 2003 10:15 am    Post subject: Reply with quote

thanks bssteph!

indeed i thought i did something wrong or the patch
was b0rked. thanks for the -R comment. w00t.
Back to top
View user's profile Send private message
bssteph
l33t
l33t


Joined: 26 Feb 2003
Posts: 651
Location: Wisconsin

PostPosted: Fri Sep 12, 2003 1:48 pm    Post subject: Reply with quote

Just booted test5-love2 on the laptop. Instantly started having the XMMS oddness, where it skips a fraction of a second while going out to disk (playing MP3s). Never had this with older kernels, to my knowledge. It only does it every now and then, but once it starts it will do it 6-10 times in the song, in order.
Perhaps the strangest part is that XMMS doesn't go into a skipfest when the system is under load. Its PR stays the same (today it's at 21) but it actually seems better under load.
load average: 2.35, 0.96, 0.42, XMMS passed all the Nautilus tests except reading /usr/portage/distfiles, where it again skipped thrice. Nautilus was a noticable degree faster this time.

system didn't have the oddness when gnome-terminal went past the bounds of the display like last time. Maybe it was just a freak accident, although it was repeatable yesterday.

windows still a bit slow when first being whipped around under load, but they accelerate to a smooth move sooner than before.

Dragging Nautilus over a transparent gnome-terminal with top running starts with the drag being slow (almost feeling like it gets behind). It normalizes in a couple seconds.

Have to go to class now. When I get back this emerge should be finished and i can try some post-emerge stuff again.

Lovechild, would you be interested in some more quantitative benchmarks, like a Contest on a couple recent kernels?
Back to top
View user's profile Send private message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Fri Sep 12, 2003 2:26 pm    Post subject: Reply with quote

Any kind of tests performed in a consistent manner on the -love kernels (preferred compared to -mm and vanilla) would be great.. I will include some scheduling latency test patches so you can all monitor regressions, if it has any interest.

But if you could wait till I get -test5-love3 out this weekend - as I've found some highly interesting patches to include - some Out of Memory tweaks, Futex speedups some compiler fixes so -love should be able to compile flawlessly (or nearly) with the ICC compilerset.

I'm still waiting for both nick v16 and a Reiser4 snapshot of the -love3 though. Nick said interesting things about high resolution timers in v16 and I'm very interested in testing that, as I think it would improve on the X redraw regressions quite a bit, or at least give us improved reaction time so it won't be felt as annoying as now.

Anyways I'm still waiting to hear if Reiser4 actually works on the love -test5 series since I had to hack it quite a bit - unless it works in it's current form, I would consider it a showstopper halting development untill namesys delivered the promised new snapshot.
Back to top
View user's profile Send private message
Wedge_
Advocate
Advocate


Joined: 08 Aug 2002
Posts: 3614
Location: Scotland

PostPosted: Fri Sep 12, 2003 2:56 pm    Post subject: Reply with quote

Lovechild wrote:
Anyways I'm still waiting to hear if Reiser4 actually works on the love -test5 series since I had to hack it quite a bit - unless it works in it's current form, I would consider it a showstopper halting development untill namesys delivered the promised new snapshot.

I have a small Reiser4 partition with a copy of the portage tree, and it works fine with test5-love1. I'm just about to try -love2.
_________________
Per Ardua Ad Astra
The Earth is the cradle of the mind, but we cannot live forever in a cradle - Konstantin E. Tsiolkovsky
Gentoo Radeon FAQ
Back to top
View user's profile Send private message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Fri Sep 12, 2003 3:26 pm    Post subject: Reply with quote

Wedge_ wrote:
Lovechild wrote:
Anyways I'm still waiting to hear if Reiser4 actually works on the love -test5 series since I had to hack it quite a bit - unless it works in it's current form, I would consider it a showstopper halting development untill namesys delivered the promised new snapshot.

I have a small Reiser4 partition with a copy of the portage tree, and it works fine with test5-love1. I'm just about to try -love2.


Could you do me a favor and REALLY hammer it good - converted all of Reiser4 to use dev_t instead of kdev since test5-love1 (hasn't changed in love2)

But if it mounts there's a good change I didn't break that nor did I break supermount (irrelevant since love2 has a proper supermount release for dev_t conversion)
Back to top
View user's profile Send private message
Wedge_
Advocate
Advocate


Joined: 08 Aug 2002
Posts: 3614
Location: Scotland

PostPosted: Fri Sep 12, 2003 3:45 pm    Post subject: Reply with quote

Lovechild wrote:
Could you do me a favor and REALLY hammer it good - converted all of Reiser4 to use dev_t instead of kdev since test5-love1 (hasn't changed in love2)

Will do. I've just rebooted with -love2, everything is still working fine.
_________________
Per Ardua Ad Astra
The Earth is the cradle of the mind, but we cannot live forever in a cradle - Konstantin E. Tsiolkovsky
Gentoo Radeon FAQ
Back to top
View user's profile Send private message
neenee
Veteran
Veteran


Joined: 20 Jul 2003
Posts: 1786

PostPosted: Fri Sep 12, 2003 3:49 pm    Post subject: Reply with quote

i did not convert to reiser4 yet, and i might not do so for
quite a while, so i hope that reiser4 will not be a must-
compile with this patchset.

apart from that possibility, i really love this patchset.
Back to top
View user's profile Send private message
nephros
Advocate
Advocate


Joined: 07 Feb 2003
Posts: 2139
Location: Graz, Austria (Europe - no kangaroos.)

PostPosted: Fri Sep 12, 2003 3:58 pm    Post subject: Reply with quote

Dear Mr. Lovechild,

have you heard about this, and ever thought about playing with it in your wonderful patchset?
_________________
Please put [SOLVED] in your topic if you are a moron.
Back to top
View user's profile Send private message
Safrax
Guru
Guru


Joined: 23 Apr 2002
Posts: 422

PostPosted: Fri Sep 12, 2003 4:11 pm    Post subject: Reply with quote

How easy would it be to add something that would allow for multiple schedulers similar to the elevator= thing? I'd love to be able to do something like scheduler=ule, nick, 0(1), or whatever at boot.
Back to top
View user's profile Send private message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Fri Sep 12, 2003 4:13 pm    Post subject: Reply with quote

nephros wrote:
Dear Mr. Lovechild,

have you heard about this, and ever thought about playing with it in your wonderful patchset?


Only a few posts away in this very thread I said I was considering that very patch for inclusion - but it failed on my computer since it was PPC only at the time so I put it in the "wait a bit" pile.

So yes, if you read the entire thread - I am considering it but I really like Nick and his scheduler ain't bad either.
Back to top
View user's profile Send private message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Fri Sep 12, 2003 4:14 pm    Post subject: Reply with quote

Safrax wrote:
How easy would it be to add something that would allow for multiple schedulers similar to the elevator= thing? I'd love to be able to do something like scheduler=ule, nick, 0(1), or whatever at boot.


Hard... I heard some of "the big brains" talking about doing selectable schedulers - but if there has been any attempts at implementation I dunno. I doubt it though.
Back to top
View user's profile Send private message
nephros
Advocate
Advocate


Joined: 07 Feb 2003
Posts: 2139
Location: Graz, Austria (Europe - no kangaroos.)

PostPosted: Fri Sep 12, 2003 4:30 pm    Post subject: Reply with quote

Lovechild wrote:
Only a few posts away in this very thread I said I was considering that very patch for inclusion
[...]
So yes, if you read the entire thread - I am considering it but I really like Nick and his scheduler ain't bad either.

Argl. Sorry.
I feel even stupider than usual now. :oops:
_________________
Please put [SOLVED] in your topic if you are a moron.
Back to top
View user's profile Send private message
Wedge_
Advocate
Advocate


Joined: 08 Aug 2002
Posts: 3614
Location: Scotland

PostPosted: Fri Sep 12, 2003 4:34 pm    Post subject: Reply with quote

Lovechild: the Reiser4 partition has handled everything I've tried so far, including copying the whole thing to another partition while writing a full size .iso onto it, with an "ls -R /usr/portage" and "find /usr/portage -iname '*'" running at the same time.
_________________
Per Ardua Ad Astra
The Earth is the cradle of the mind, but we cannot live forever in a cradle - Konstantin E. Tsiolkovsky
Gentoo Radeon FAQ
Back to top
View user's profile Send private message
ahr
n00b
n00b


Joined: 14 Jul 2002
Posts: 31
Location: Reykjavik, Iceland

PostPosted: Fri Sep 12, 2003 6:28 pm    Post subject: Reply with quote

Lovechild wrote:
nephros wrote:
Dear Mr. Lovechild,

have you heard about this, and ever thought about playing with it in your wonderful patchset?


Only a few posts away in this very thread I said I was considering that very patch for inclusion - but it failed on my computer since it was PPC only at the time so I put it in the "wait a bit" pile.

So yes, if you read the entire thread - I am considering it but I really like Nick and his scheduler ain't bad either.


There is a i386 version there now. I tried it earler and it my laptop hard freezes when testing it. It happens random, I cannot duplicate it. And I am no kernel-hacker, I have no idea how to submit stuff so the developer can look at the problem.

I'm running an Centrino laptop. HP-Compaq nx7000. Running 2.6-test4 without a glitch here. Haven't tested unpatched test5 yet.
Back to top
View user's profile Send private message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Fri Sep 12, 2003 7:00 pm    Post subject: Reply with quote

Wedge_ wrote:
Lovechild: the Reiser4 partition has handled everything I've tried so far, including copying the whole thing to another partition while writing a full size .iso onto it, with an "ls -R /usr/portage" and "find /usr/portage -iname '*'" running at the same time.


You are really going to love me now.. Reiser4 was released in a new snapshot, and it breaks the disk format so you will have to compile new tools and libs Reiser4 and recreate your partition..

oh and yes - I'm testing love3 now - I handmerged the ULE port, I did this BEFORE reading the comment about random hangs, but now I'll test it.

ULE I hear is a port of the O(1) scheduler to FreeBSD and then they improved the interactive handling - so it's really a port of a port now, nastiness - but FreeBSD and Linux having the same scheduler could be good for comparing the two kernels.
Back to top
View user's profile Send private message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Fri Sep 12, 2003 7:06 pm    Post subject: Reply with quote

The ULE stuff seems.... interesting :

Unable to handle kernel NULL pointer dereference at virtual address 00000078
printing eip:
c029dfe2
*pde = 00000000
Oops: 0000 [#1]
CPU: 0
EIP: 0060:[<c029dfe2>] Not tainted
EFLAGS: 00010297
EIP is at vsnprintf+0x302/0x450
eax: 00000078 ebx: 0000000a ecx: 00000078 edx: fffffffe
esi: dee9c052 edi: 00000000 ebp: ffffffff esp: de9b3e60
ds: 007b es: 007b ss: 0068
Process ps (pid: 2566, threadinfo=de9b2000 task=df24cc80)
Stack: dee9c047 ffffffff 00000078 00000000 0000000a ffffffff 00000002 00000002
ffffffff ffffffff dee9c00b de9b2000 c04235de c152d8c0 c029e187 dee9c00b
21163ff5 c04296bb de9b3edc c029e1af dee9c00b c0429680 de9b3ec8 c0185bef
Call Trace:
[<c029e187>] vsprintf+0x27/0x30
[<c029e1af>] sprintf+0x1f/0x30
[<c0185bef>] proc_pid_status+0x1bf/0x4a0
[<c01833c4>] proc_info_read+0x54/0x140
[<c0155242>] filp_open+0x62/0x70
[<c0156028>] vfs_read+0xb8/0x130
[<c01562d2>] sys_read+0x42/0x70
[<c01092c9>] sysenter_past_esp+0x52/0x71

Code: 80 38 00 74 07 40 4a 83 fa ff 75 f4 29 c8 83 e7 10 89 c3 75
<6>note: ps[2566] exited with preempt_count 1
Unable to handle kernel NULL pointer dereference at virtual address 00000078
printing eip:
c029dfe2
*pde = 00000000
Oops: 0000 [#2]
CPU: 0
EIP: 0060:[<c029dfe2>] Not tainted
EFLAGS: 00010297
EIP is at vsnprintf+0x302/0x450
eax: 00000078 ebx: 0000000a ecx: 00000078 edx: fffffffe
esi: df045052 edi: 00000000 ebp: ffffffff esp: df76de60
ds: 007b es: 007b ss: 0068
Process ps (pid: 2758, threadinfo=df76c000 task=dfd080c0)
Stack: df045047 ffffffff 00000078 00000000 0000000a ffffffff 00000002 00000002
ffffffff ffffffff df04500b df76c000 c04235de c152d8c0 c029e187 df04500b
20fbaff5 c04296bb df76dedc c029e1af df04500b c0429680 df76dec8 c0185bef
Call Trace:
[<c029e187>] vsprintf+0x27/0x30
[<c029e1af>] sprintf+0x1f/0x30
[<c0185bef>] proc_pid_status+0x1bf/0x4a0
[<c01833c4>] proc_info_read+0x54/0x140
[<c0155242>] filp_open+0x62/0x70
[<c0156028>] vfs_read+0xb8/0x130
[<c01562d2>] sys_read+0x42/0x70
[<c01092c9>] sysenter_past_esp+0x52/0x71

Code: 80 38 00 74 07 40 4a 83 fa ff 75 f4 29 c8 83 e7 10 89 c3 75
<6>note: ps[2758] exited with preempt_count 1




But DAMN it's smooth at first glance

-edit-

Strike that, it's absolutely crap under load -- absolutely crap, windows are choppy - it's just nasty.

-edit2-
This scheduler makes me sick - the music chopping is back even without any load... bad.. very bad.. only one word can describe it... *cough* crapular *cough*
Back to top
View user's profile Send private message
Wedge_
Advocate
Advocate


Joined: 08 Aug 2002
Posts: 3614
Location: Scotland

PostPosted: Fri Sep 12, 2003 7:19 pm    Post subject: Reply with quote

Lovechild wrote:
You are really going to love me now.. Reiser4 was released in a new snapshot, and it breaks the disk format so you will have to compile new tools and libs Reiser4 and recreate your partition..

Yay :P
_________________
Per Ardua Ad Astra
The Earth is the cradle of the mind, but we cannot live forever in a cradle - Konstantin E. Tsiolkovsky
Gentoo Radeon FAQ
Back to top
View user's profile Send private message
Safrax
Guru
Guru


Joined: 23 Apr 2002
Posts: 422

PostPosted: Fri Sep 12, 2003 10:01 pm    Post subject: Reply with quote

Lovechild wrote:
But if you could wait till I get -test5-love3 out this weekend - as I've found some highly interesting patches to include - some Out of Memory tweaks, Futex speedups some compiler fixes so -love should be able to compile flawlessly (or nearly) with the ICC compilerset.


OMFG, a kernel compiled by icc would be amazing for my p4-m. I can't wait to attempt it!
Back to top
View user's profile Send private message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Fri Sep 12, 2003 10:13 pm    Post subject: Reply with quote

Safrax wrote:
Lovechild wrote:
But if you could wait till I get -test5-love3 out this weekend - as I've found some highly interesting patches to include - some Out of Memory tweaks, Futex speedups some compiler fixes so -love should be able to compile flawlessly (or nearly) with the ICC compilerset.


OMFG, a kernel compiled by icc would be amazing for my p4-m. I can't wait to attempt it!


Lemme just back out the ULE scheduler, which has to be done by hand (crappy online archive of lkml doesn't format the patch right - and it wasn't in alignment anyways) - and put the lastest Nick scheduler back in.

I'm considering backing out the OOM addons as we really don't need to be able to set the OOM killer settings at compile time on a desktop machine - mainly I think it sounded cool. The futex tweaks have shown upto 6% speed increase in futex converted programs - which is quite cool. Futexes rock !

And yes ICC should be able to compile the kernel, so that might be worth a good few cycles - despite it being hand tweaked by the kernel hackers.

I will do the work on -love3 tomorrow - right now... Bedtime, I have a support call in the morning.
Back to top
View user's profile Send private message
bssteph
l33t
l33t


Joined: 26 Feb 2003
Posts: 651
Location: Wisconsin

PostPosted: Sat Sep 13, 2003 5:58 am    Post subject: Reply with quote

I know, you haven't released love3 yet, but I couldn't resist.

http://bssteph.irtonline.org/linux/testing/2.6.0-test5_vanilla_mm1-O20.1_love.txt
contest of 2.6.0-test5 and it patched with -mm1+O20.1, -love1, -love2

To no real surprise -mm1 and -love2 lead the pack, but I'm surprised by how much worse -love1 is in the io_* tests. -mm1 is whooped by -love in dbench but judging by the man page this probably has a lot to do with the cfq elevator.

If you think I should try the contests in a different manner/using other kernels, let me know. I'll do them when I do love3, but there's a good chance that won't be until Sunday afternoon. In retrospect I probably should have done elevator=as on the love kernels but I think the only one it'd have a huge affect on is dbench.... right?

Now time for me to sleep.
Back to top
View user's profile Send private message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Sat Sep 13, 2003 7:48 am    Post subject: Reply with quote

Love gets whooped in the IO tests because -mm1 and vanilla uses the AS scheduler which is blazing fast for IO - where as -love uses CFQ per default, which aims for complete fairness.

If you rerun the test with elevator=as appended to the -love kernel you should see improvements in that area.

The tradeoff is for fairness, not for speed - raw IO speed gains you little on a desktop machine, whereas fairness is a better idea in some cases - to ensure every process has equal right to IO access, so mplayer never skips a frame, xmms never gets to little data - etc.

-love is a desktop kernel it's not really meant to do good in a test like contest - it's meant to do good in interactive tests.

I noticed that -love generally has better CPU utilisation than both vanilla and -mm - this is good, very good.

But looking at the tests you will notice that -love consistently has as high or higher CPU utilisation, and lower time usage. especially note the dbench result where -love whoops everyones ass... hehe.

In conclusion I would say that aside the Process_load and mem_load tests -love comes out on top or on par in every test - not bad, not bad at all.
Back to top
View user's profile Send private message
ed0n
l33t
l33t


Joined: 23 Apr 2003
Posts: 638
Location: Prishtine/Kosove

PostPosted: Sat Sep 13, 2003 7:55 am    Post subject: Reply with quote

Those are the sources that I needed thank you Lovechild ;)
Back to top
View user's profile Send private message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Sat Sep 13, 2003 8:29 am    Post subject: Reply with quote

ed0n wrote:
Those are the sources that I needed thank you Lovechild ;)


I recall you have / on Reiser4 - so DON'T install love3 on that partition or it will go boom...
Back to top
View user's profile Send private message
Lovechild
Advocate
Advocate


Joined: 17 May 2002
Posts: 2858
Location: Århus, Denmark

PostPosted: Sat Sep 13, 2003 1:42 pm    Post subject: Reply with quote

it compiles... SHIP IT !!

http://userportal.iha.dk/~01876/2.6.0-test5-love3.tar.bz2

includes:

Nick v15
Supermount 2.02
Futexes tweaks
Useless OOM stuff with a cool sounding name
CFQ
Reiser4
Vesafb patch
ICC patch
GCC 3.4 patch
everything I forgot

WARNING !!!

:!: :!: :!: :!: :!: :!:

The data format in Reiser4 has changed - it's important that you build the newly released tools and libs - and recreate your Reiser4 partitions.

:!: :!: :!: :!: :!: :!:
Back to top
View user's profile Send private message
Display posts from previous:   
Reply to topic    Gentoo Forums Forum Index Off the Wall All times are GMT
Goto page Previous  1, 2, 3, 4, 5, 6, 7  Next
Page 3 of 7

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum