2004-01-11 20:08:12

by ciaby

[permalink] [raw]
Subject: kmail slowdown on 2.6.* +reiserFS (v3)

I all!
I've recently upgraded from 2.4 to 2.6 and I've noticed a strange thing:
on the 2.4 kernel, kmail run decently (i've an old k6-200).
On the 2.6 kernel, kmail slowdown and take a very long time to read a mailbox.
I think something changed in the reiserFS during this time...
I'm not the only experiencing this problem, read this:
http://kerneltrap.org/node/view/1844
Please CC me if there's some reply.
tnx all for the good work! :-)

Bye
--
-----> \|/Ciaby\|/ <------
! E-mail:[email protected] !
! ICQ:59493039 !
! SpinHacker404 HackLab !
! http://www.ecn.org/sh404 !
gpg --keyserver autistici.org --recv-key 42F3C487


2004-01-12 02:27:44

by Andrew Morton

[permalink] [raw]
Subject: Re: kmail slowdown on 2.6.* +reiserFS (v3)

Ciaby <[email protected]> wrote:
>
> I all!
> I've recently upgraded from 2.4 to 2.6 and I've noticed a strange thing:
> on the 2.4 kernel, kmail run decently (i've an old k6-200).
> On the 2.6 kernel, kmail slowdown and take a very long time to read a mailbox.
> I think something changed in the reiserFS during this time...
> I'm not the only experiencing this problem, read this:
> http://kerneltrap.org/node/view/1844

A buglet in kmail was tripped up by some optimisations which went into
reiserfs.

Upgrading kmail should fix it up. Or mount the reiserfs filesystems with
the `nolargeio=1' mount option.

2004-01-12 13:19:52

by Samium Gromoff

[permalink] [raw]
Subject: Re: kmail slowdown on 2.6.* +reiserFS (v3)


> I all!
> I've recently upgraded from 2.4 to 2.6 and I've noticed a strange thing:
> on the 2.4 kernel, kmail run decently (i've an old k6-200).
> On the 2.6 kernel, kmail slowdown and take a very long time to read a mailbox.
> I think something changed in the reiserFS during this time...
> I'm not the only experiencing this problem, read this:
> http://kerneltrap.org/node/view/1844
> Please CC me if there's some reply.
> tnx all for the good work! :-)

How much RAM do you have?

This might be the 2.6 VM regression hitting you.

> Bye
> --
> -----> \|/Ciaby\|/ <------
> ! E-mail:[email protected] !
> ! ICQ:59493039 !
> ! SpinHacker404 HackLab !
> ! http://www.ecn.org/sh404 !
> gpg --keyserver autistici.org --recv-key 42F3C487

regards, Samium Gromoff


2004-01-14 13:19:24

by Chris Mason

[permalink] [raw]
Subject: Re: kmail slowdown on 2.6.* +reiserFS (v3)

On Sun, 2004-01-11 at 21:27, Andrew Morton wrote:
> Ciaby <[email protected]> wrote:
> >
> > I all!
> > I've recently upgraded from 2.4 to 2.6 and I've noticed a strange thing:
> > on the 2.4 kernel, kmail run decently (i've an old k6-200).
> > On the 2.6 kernel, kmail slowdown and take a very long time to read a mailbox.
> > I think something changed in the reiserFS during this time...
> > I'm not the only experiencing this problem, read this:
> > http://kerneltrap.org/node/view/1844
>
> A buglet in kmail was tripped up by some optimisations which went into
> reiserfs.
>
> Upgrading kmail should fix it up. Or mount the reiserfs filesystems with
> the `nolargeio=1' mount option.

Actually, we've hit other problems with v3 largeio, it can confuse rpm
badly. The real bug is apparently in bdb, the larger io size suggested
by the filesystem lead bdb to corrupt its own files. I spent some time
neck deep in the db code but couldn't track the problem down.

I seem to remember the XFS folks hitting exactly the same bug.

Hans, can I talk you into having v3 export an io size of 4k to userspace
again? Applications that send large ios would still use Oleg's
optimized file write paths.

-chris


2004-01-14 13:32:20

by Hans Reiser

[permalink] [raw]
Subject: Re: kmail slowdown on 2.6.* +reiserFS (v3)

Chris Mason wrote:

>On Sun, 2004-01-11 at 21:27, Andrew Morton wrote:
>
>
>>Ciaby <[email protected]> wrote:
>>
>>
>>>I all!
>>>I've recently upgraded from 2.4 to 2.6 and I've noticed a strange thing:
>>>on the 2.4 kernel, kmail run decently (i've an old k6-200).
>>>On the 2.6 kernel, kmail slowdown and take a very long time to read a mailbox.
>>>I think something changed in the reiserFS during this time...
>>>I'm not the only experiencing this problem, read this:
>>>http://kerneltrap.org/node/view/1844
>>>
>>>
>>A buglet in kmail was tripped up by some optimisations which went into
>>reiserfs.
>>
>>Upgrading kmail should fix it up. Or mount the reiserfs filesystems with
>>the `nolargeio=1' mount option.
>>
>>
>
>Actually, we've hit other problems with v3 largeio, it can confuse rpm
>badly. The real bug is apparently in bdb, the larger io size suggested
>by the filesystem lead bdb to corrupt its own files. I spent some time
>neck deep in the db code but couldn't track the problem down.
>
>I seem to remember the XFS folks hitting exactly the same bug.
>
>Hans, can I talk you into having v3 export an io size of 4k to userspace
>again? Applications that send large ios would still use Oleg's
>optimized file write paths.
>
>-chris
>
>
>
>
>
>
why is it you don't want to "fix" bdb to lie to itself about the result
of statfs?

--
Hans


2004-01-14 13:40:45

by Chris Mason

[permalink] [raw]
Subject: Re: kmail slowdown on 2.6.* +reiserFS (v3)

On Wed, 2004-01-14 at 08:32, Hans Reiser wrote:
> Chris Mason wrote:
>
> >On Sun, 2004-01-11 at 21:27, Andrew Morton wrote:
> >
> >
> >>Ciaby <[email protected]> wrote:
> >>
> >>
> >>>I all!
> >>>I've recently upgraded from 2.4 to 2.6 and I've noticed a strange thing:
> >>>on the 2.4 kernel, kmail run decently (i've an old k6-200).
> >>>On the 2.6 kernel, kmail slowdown and take a very long time to read a mailbox.
> >>>I think something changed in the reiserFS during this time...
> >>>I'm not the only experiencing this problem, read this:
> >>>http://kerneltrap.org/node/view/1844
> >>>
> >>>
> >>A buglet in kmail was tripped up by some optimisations which went into
> >>reiserfs.
> >>
> >>Upgrading kmail should fix it up. Or mount the reiserfs filesystems with
> >>the `nolargeio=1' mount option.
> >>
> >>
> >
> >Actually, we've hit other problems with v3 largeio, it can confuse rpm
> >badly. The real bug is apparently in bdb, the larger io size suggested
> >by the filesystem lead bdb to corrupt its own files. I spent some time
> >neck deep in the db code but couldn't track the problem down.
> >
> >I seem to remember the XFS folks hitting exactly the same bug.
> >
> >Hans, can I talk you into having v3 export an io size of 4k to userspace
> >again? Applications that send large ios would still use Oleg's
> >optimized file write paths.
> >
> why is it you don't want to "fix" bdb to lie to itself about the result
> of statfs?

Because I'm worried that bdb isn't the only app having problems ;-) And
it's important enough as a legacy app that I don't want to tell everyone
they must upgrade to some hacked bdb version in order for v3 to work
under 2.6.x

-chris