Mailing List Archive: JFS on non-intel [ Was O2 Sound and Misc Stuff]

JFS on non-intel [ Was O2 Sound and Misc Stuff]

Feb 9, 2007, 11:39 AM

Post #1 of 4 (3645 views)

I'm interested in your data point. Do you happen to remember about which
kernel that was? There was some general badness that affected multiple
file systems in late 2.6.13 on into 2.6.14 or so in the way that you
describe. Not sure they ever really knew what the smoking gun was.

Used JFS for about 6 months with the gentoo mips 2.6.13 kernel and did
not find any issues. Briefly with Sparc32, but don't remember
which kernel. On Intel, I've done extensive regressions with it.

If anyone is interested, I have a thrasher script written specifically for
this purpose. It's a multi-threaded ruby script that creates a randomized
directory tree, with random files, containing random data. A CRC check is
kept on each file. The threads run in parallel rewriting data in the
middle of files, truncating, resizing, creating voids, forcing the file
system into writing multiple extents, exercising tail packing, unlinking
files with open handles, etc...

-S-

On Fri, 9 Feb 2007, Alex Deucher wrote:

> On 2/9/07, J. Scott Kasten <jscottkasten@yahoo.com> wrote:

> Be careful. I tried to use JFS on some sparc boxes, and ran into some
> subtle bugs that no one seemed to be able to solve that led to
> filesystem corruption: things like disappearing/reappearing files and
> directories. On the other hand I've had no problems with JFS on AMD64
> or x86.
>
> Alex
>
--
gentoo-mips@gentoo.org mailing list

Re: JFS on non-intel [ Was O2 Sound and Misc Stuff] [ In reply to ]

alexdeucher at gmail

Feb 9, 2007, 12:05 PM

Post #2 of 4 (3478 views)

Permalink

On 2/9/07, J. Scott Kasten <jscottkasten@yahoo.com> wrote:
>
> I'm interested in your data point. Do you happen to remember about which
> kernel that was? There was some general badness that affected multiple
> file systems in late 2.6.13 on into 2.6.14 or so in the way that you
> describe. Not sure they ever really knew what the smoking gun was.
>
> Used JFS for about 6 months with the gentoo mips 2.6.13 kernel and did
> not find any issues. Briefly with Sparc32, but don't remember
> which kernel. On Intel, I've done extensive regressions with it.
>
> If anyone is interested, I have a thrasher script written specifically for
> this purpose. It's a multi-threaded ruby script that creates a randomized
> directory tree, with random files, containing random data. A CRC check is
> kept on each file. The threads run in parallel rewriting data in the
> middle of files, truncating, resizing, creating voids, forcing the file
> system into writing multiple extents, exercising tail packing, unlinking
> files with open handles, etc...
>

http://sourceforge.net/mailarchive/forum.php?thread_id=7852301&forum_id=43911
http://www.mail-archive.com/sparclinux@vger.kernel.org/msg00333.html

Alex

> -S-
>
> On Fri, 9 Feb 2007, Alex Deucher wrote:
>
> > On 2/9/07, J. Scott Kasten <jscottkasten@yahoo.com> wrote:
>
> > Be careful. I tried to use JFS on some sparc boxes, and ran into some
> > subtle bugs that no one seemed to be able to solve that led to
> > filesystem corruption: things like disappearing/reappearing files and
> > directories. On the other hand I've had no problems with JFS on AMD64
> > or x86.
> >
> > Alex
> >
> --
> gentoo-mips@gentoo.org mailing list
>
>
--
gentoo-mips@gentoo.org mailing list

Re: JFS on non-intel [ Was O2 Sound and Misc Stuff] [ In reply to ]

jscottkasten at yahoo

Feb 9, 2007, 2:39 PM

Post #3 of 4 (3486 views)

Permalink

Alex, thanks. I read through it in detail. That finally triggered my
memory. There were multiple issues that affected 2.6.12 - 2.6.14.

In your case, it probably wasn't JFS specifically, but most likely an
interaction between JFS and the LVM layer. I never tried that feature.
This affected 2.6.12 and 2.6.13

http://www.arcknowledge.com/gmane.comp.file-systems.jfs.general/2005-10/msg00020.html

Here is also an interesting LWN article about the 4k page issue that came
out at the same time and how it affected device_mapper, LVM and such
things. (See the second section.)

http://lwn.net/Articles/149977/

Here's a complaint about 2.6.13 and 2.6.14 on AMD with ext3. This is just
an example of other complaints that were comming in.

http://lkml.org/lkml/2005/09/27/62

In short, there were multiple sources of corruption at that time and you
could have been bitten by any of them. Hopefully things are more settled
now, but I certainly will watch out for any issues.

-S-

On Fri, 9 Feb 2007, Alex Deucher wrote:

> On 2/9/07, J. Scott Kasten <jscottkasten@yahoo.com> wrote:
>>
>> I'm interested in your data point. Do you happen to remember about which
>> kernel that was? There was some general badness that affected multiple
>> file systems in late 2.6.13 on into 2.6.14 or so in the way that you
>> describe. Not sure they ever really knew what the smoking gun was.
>
> http: //sourceforge.net/mailarchive/forum.php?thread_id=7852301&forum_id=43911
> http: //www.mail-archive.com/sparclinux@vger.kernel.org/msg00333.html
>
> Alex
>
--
gentoo-mips@gentoo.org mailing list

Re: JFS on non-intel [ Was O2 Sound and Misc Stuff] [ In reply to ]

alexdeucher at gmail

Feb 9, 2007, 3:30 PM

Post #4 of 4 (3482 views)

Permalink

On 2/9/07, J. Scott Kasten <jscottkasten@yahoo.com> wrote:
>
> Alex, thanks. I read through it in detail. That finally triggered my
> memory. There were multiple issues that affected 2.6.12 - 2.6.14.
>
> In your case, it probably wasn't JFS specifically, but most likely an
> interaction between JFS and the LVM layer. I never tried that feature.
> This affected 2.6.12 and 2.6.13
>
> http://www.arcknowledge.com/gmane.comp.file-systems.jfs.general/2005-10/msg00020.html
>
> Here is also an interesting LWN article about the 4k page issue that came
> out at the same time and how it affected device_mapper, LVM and such
> things. (See the second section.)
>
> http://lwn.net/Articles/149977/
>
> Here's a complaint about 2.6.13 and 2.6.14 on AMD with ext3. This is just
> an example of other complaints that were comming in.
>
> http://lkml.org/lkml/2005/09/27/62
>
> In short, there were multiple sources of corruption at that time and you
> could have been bitten by any of them. Hopefully things are more settled
> now, but I certainly will watch out for any issues.

Thanks for the update! those indeed look like they may have been the
cause of my problems. Too bad I can't test that set up again to see
if it works today.

Alex

>
> -S-
>
> On Fri, 9 Feb 2007, Alex Deucher wrote:
>
> > On 2/9/07, J. Scott Kasten <jscottkasten@yahoo.com> wrote:
> >>
> >> I'm interested in your data point. Do you happen to remember about which
> >> kernel that was? There was some general badness that affected multiple
> >> file systems in late 2.6.13 on into 2.6.14 or so in the way that you
> >> describe. Not sure they ever really knew what the smoking gun was.
> >
> > http: //sourceforge.net/mailarchive/forum.php?thread_id=7852301&forum_id=43911
> > http: //www.mail-archive.com/sparclinux@vger.kernel.org/msg00333.html
> >
> > Alex
> >
> --
> gentoo-mips@gentoo.org mailing list
>
>
--
gentoo-mips@gentoo.org mailing list