* Re: Interactive Usage of 2.6.0.test1 worse than 2.4.21
2003-08-05 5:04 Interactive Usage of 2.6.0.test1 worse than 2.4.21 Martin Konold
@ 2003-08-05 5:16 ` William Lee Irwin III
2003-08-05 6:26 ` Andrew Morton
` (2 subsequent siblings)
3 siblings, 0 replies; 11+ messages in thread
From: William Lee Irwin III @ 2003-08-05 5:16 UTC (permalink / raw)
To: Martin Konold; +Cc: linux-kernel
On Tue, Aug 05, 2003 at 07:04:22AM +0200, Martin Konold wrote:
> when using 2.6.0.test1 on a high end laptop (P-IV 2.2 GHz, 1GB RAM) I notice
> very significant slowdown in interactive usage compared to 2.4.21.
> The difference is most easily seen when switching folders in kmail. While
> 2.4.21 is instantaneous 2.6.0.test1 shows the clock for about 2-3 seconds.
> I am using maildir folders on reiserfs.
> Can anyone verify this behaviour?
> What other information do you need?
CPU profiles, e.g.
readprofile -n -m /boot/System.map-`uname -r` | sort -rn -k 1,1 | head -25
Also logs of vmstat 1.
-- wli
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Interactive Usage of 2.6.0.test1 worse than 2.4.21
2003-08-05 5:04 Interactive Usage of 2.6.0.test1 worse than 2.4.21 Martin Konold
2003-08-05 5:16 ` William Lee Irwin III
@ 2003-08-05 6:26 ` Andrew Morton
2003-08-05 8:16 ` Oleg Drokin
2003-08-05 22:43 ` Peter Chubb
2003-08-05 7:23 ` Felipe Alfaro Solana
2003-08-05 12:52 ` Bernd Schubert
3 siblings, 2 replies; 11+ messages in thread
From: Andrew Morton @ 2003-08-05 6:26 UTC (permalink / raw)
To: Martin Konold; +Cc: linux-kernel
Martin Konold <martin.konold@erfrakon.de> wrote:
>
> Hi,
>
> when using 2.6.0.test1 on a high end laptop (P-IV 2.2 GHz, 1GB RAM) I notice
> very significant slowdown in interactive usage compared to 2.4.21.
>
> The difference is most easily seen when switching folders in kmail. While
> 2.4.21 is instantaneous 2.6.0.test1 shows the clock for about 2-3 seconds.
>
> I am using maildir folders on reiserfs.
There is a bug in old kmail versions wherein they do silly things if the
filesystem alleges that its optimum I/O size is much larger than 4k.
2.6's reiserfs tell applications that its optimum IO size is 128k, and the
bug bites.
Try mounting your reiserfs filesystems with the "nolargeio" option.
A `mount -o remount,nolargeio' will probably work too.
Please test that, send a report, and if it fixes it, upgrade your kmail.
Thanks.
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Interactive Usage of 2.6.0.test1 worse than 2.4.21
2003-08-05 6:26 ` Andrew Morton
@ 2003-08-05 8:16 ` Oleg Drokin
2003-08-05 22:43 ` Peter Chubb
1 sibling, 0 replies; 11+ messages in thread
From: Oleg Drokin @ 2003-08-05 8:16 UTC (permalink / raw)
To: Andrew Morton; +Cc: Martin Konold, linux-kernel
Hello!
On Mon, Aug 04, 2003 at 11:26:54PM -0700, Andrew Morton wrote:
> Try mounting your reiserfs filesystems with the "nolargeio" option.
> A `mount -o remount,nolargeio' will probably work too.
nolargeio requires an argument, so it should look like
mount -o remount,nolargeio=1
Bye,
Oleg
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Interactive Usage of 2.6.0.test1 worse than 2.4.21
2003-08-05 6:26 ` Andrew Morton
2003-08-05 8:16 ` Oleg Drokin
@ 2003-08-05 22:43 ` Peter Chubb
2003-08-05 23:36 ` Andrew Morton
2003-08-06 0:55 ` Nick Piggin
1 sibling, 2 replies; 11+ messages in thread
From: Peter Chubb @ 2003-08-05 22:43 UTC (permalink / raw)
To: Andrew Morton; +Cc: Martin Konold, linux-kernel
>>>>> "Andrew" == Andrew Morton <akpm@osdl.org> writes:
Andrew> Martin Konold <martin.konold@erfrakon.de> wrote:
>> Hi,
>>
>> when using 2.6.0.test1 on a high end laptop (P-IV 2.2 GHz, 1GB RAM)
>> I notice very significant slowdown in interactive usage compared to
>> 2.4.21.
>>
>> The difference is most easily seen when switching folders in
>> kmail. While 2.4.21 is instantaneous 2.6.0.test1 shows the clock
>> for about 2-3 seconds.
>>
I see the same problem, and I'm using XFS. Booting with
elevator=deadline fixed it for me. The anticipatory scheduler hurts
if you have a disc optimised for low power consumption, not speed.
--
Dr Peter Chubb http://www.gelato.unsw.edu.au peterc AT gelato.unsw.edu.au
You are lost in a maze of BitKeeper repositories, all slightly different.
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Interactive Usage of 2.6.0.test1 worse than 2.4.21
2003-08-05 22:43 ` Peter Chubb
@ 2003-08-05 23:36 ` Andrew Morton
2003-08-06 0:55 ` Nick Piggin
1 sibling, 0 replies; 11+ messages in thread
From: Andrew Morton @ 2003-08-05 23:36 UTC (permalink / raw)
To: Peter Chubb; +Cc: martin.konold, linux-kernel
Peter Chubb <peter@chubb.wattle.id.au> wrote:
>
> >> when using 2.6.0.test1 on a high end laptop (P-IV 2.2 GHz, 1GB RAM)
> >> I notice very significant slowdown in interactive usage compared to
> >> 2.4.21.
> >>
> >> The difference is most easily seen when switching folders in
> >> kmail. While 2.4.21 is instantaneous 2.6.0.test1 shows the clock
> >> for about 2-3 seconds.
> >>
>
> I see the same problem, and I'm using XFS. Booting with
> elevator=deadline fixed it for me. The anticipatory scheduler hurts
> if you have a disc optimised for low power consumption, not speed.
Do you have a specific set of steps with which to reproduce this?
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Interactive Usage of 2.6.0.test1 worse than 2.4.21
2003-08-05 22:43 ` Peter Chubb
2003-08-05 23:36 ` Andrew Morton
@ 2003-08-06 0:55 ` Nick Piggin
1 sibling, 0 replies; 11+ messages in thread
From: Nick Piggin @ 2003-08-06 0:55 UTC (permalink / raw)
To: Peter Chubb; +Cc: Andrew Morton, Martin Konold, linux-kernel
Peter Chubb wrote:
>>>>>>"Andrew" == Andrew Morton <akpm@osdl.org> writes:
>>>>>>
>
>Andrew> Martin Konold <martin.konold@erfrakon.de> wrote:
>
>>>Hi,
>>>
>>>when using 2.6.0.test1 on a high end laptop (P-IV 2.2 GHz, 1GB RAM)
>>>I notice very significant slowdown in interactive usage compared to
>>>2.4.21.
>>>
>>>The difference is most easily seen when switching folders in
>>>kmail. While 2.4.21 is instantaneous 2.6.0.test1 shows the clock
>>>for about 2-3 seconds.
>>>
>>>
>
>I see the same problem, and I'm using XFS. Booting with
>elevator=deadline fixed it for me. The anticipatory scheduler hurts
>if you have a disc optimised for low power consumption, not speed.
>
>
I don't think this generalisation is really fair. All hard disks
have the same basic properties which AS exploits. There seems to
be something going wrong though.
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Interactive Usage of 2.6.0.test1 worse than 2.4.21
2003-08-05 5:04 Interactive Usage of 2.6.0.test1 worse than 2.4.21 Martin Konold
2003-08-05 5:16 ` William Lee Irwin III
2003-08-05 6:26 ` Andrew Morton
@ 2003-08-05 7:23 ` Felipe Alfaro Solana
2003-08-05 12:52 ` Bernd Schubert
3 siblings, 0 replies; 11+ messages in thread
From: Felipe Alfaro Solana @ 2003-08-05 7:23 UTC (permalink / raw)
To: Martin Konold; +Cc: LKML
On Tue, 2003-08-05 at 07:04, Martin Konold wrote:
> Hi,
>
> when using 2.6.0.test1 on a high end laptop (P-IV 2.2 GHz, 1GB RAM) I notice
> very significant slowdown in interactive usage compared to 2.4.21.
Please, upgrade to the latest 2.6.0-test kernel, as there are a lot of
people working on the CPU scheduler and interactivity. As of this mail,
it's 2.6.0-test2-bk4. If you prefer, you can also test 2.6.0-test2-mm4
(Andrew Morton patches on top of 2.6.0-test2).
This way, you can help us in improving the interactive feeling of future
2.6 kernels.
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Interactive Usage of 2.6.0.test1 worse than 2.4.21
2003-08-05 5:04 Interactive Usage of 2.6.0.test1 worse than 2.4.21 Martin Konold
` (2 preceding siblings ...)
2003-08-05 7:23 ` Felipe Alfaro Solana
@ 2003-08-05 12:52 ` Bernd Schubert
2003-08-05 13:37 ` Rahul Karnik
3 siblings, 1 reply; 11+ messages in thread
From: Bernd Schubert @ 2003-08-05 12:52 UTC (permalink / raw)
To: linux-kernel
On Tuesday 05 August 2003 07:04, Martin Konold wrote:
> Hi,
>
> when using 2.6.0.test1 on a high end laptop (P-IV 2.2 GHz, 1GB RAM) I
> notice very significant slowdown in interactive usage compared to 2.4.21.
>
> The difference is most easily seen when switching folders in kmail. While
> 2.4.21 is instantaneous 2.6.0.test1 shows the clock for about 2-3 seconds.
>
> I am using maildir folders on reiserfs.
>
> Can anyone verify this behaviour?
>
Yes, I can definitely verify this, its not only related to kde/kmail, all
other application are affected as well. Btw, I already upgraded to
2.6.0-test2.
Also, the slowdown seems to be related to hd-accessing/caching. My root-fs hd
makes rather loud noises on accessing it -- with 2.6.0-testX the frequency of
disk-accessing and so also the noise-level has dramatically increased
compared to 2.4.x
So following the advices, I will try bk4 and mm4 and also will do the
>CPU profiles, e.g.
>readprofile -n -m /boot/System.map-`uname -r` | sort -rn -k 1,1 | head -25
>Also logs of vmstat 1.
stuff.
Regards,
Bernd
--
Bernd Schubert
Physikalisch Chemisches Institut / Theoretische Chemie
Universität Heidelberg
INF 229
69120 Heidelberg
e-mail: bernd.schubert@pci.uni-heidelberg.de
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Interactive Usage of 2.6.0.test1 worse than 2.4.21
2003-08-05 12:52 ` Bernd Schubert
@ 2003-08-05 13:37 ` Rahul Karnik
2003-08-05 21:29 ` Martin Konold
0 siblings, 1 reply; 11+ messages in thread
From: Rahul Karnik @ 2003-08-05 13:37 UTC (permalink / raw)
To: Bernd Schubert; +Cc: linux-kernel
If disk is involved, your problem might simply be the incorrect
readahead value. Try "hdparm -a 512".
http://marc.theaimsgroup.com/?l=linux-kernel&m=105830624016066&w=2
-Rahul
--
Rahul Karnik
rahul@genebrew.com
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Interactive Usage of 2.6.0.test1 worse than 2.4.21
2003-08-05 13:37 ` Rahul Karnik
@ 2003-08-05 21:29 ` Martin Konold
0 siblings, 0 replies; 11+ messages in thread
From: Martin Konold @ 2003-08-05 21:29 UTC (permalink / raw)
To: Rahul Karnik; +Cc: linux-kernel
Am Dienstag, 5. August 2003 15:37 schrieb Rahul Karnik:
Hi,
> If disk is involved, your problem might simply be the incorrect
> readahead value. Try "hdparm -a 512".
This improved the situation noticeably while still beeing slower than 2.4.21.
I will test more recent kernels later tonight.
Regards,
-- martin
Dipl.-Phys. Martin Konold
e r f r a k o n
Erlewein, Frank, Konold & Partner - Beratende Ingenieure und Physiker
Nobelstrasse 15, 70569 Stuttgart, Germany
fon: 0711 67400963, fax: 0711 67400959
email: martin.konold@erfrakon.de
^ permalink raw reply [flat|nested] 11+ messages in thread