[mythtv-users] IOBOUND errors after kernel upgrade from 2.6.31 to 2.6.32

[Tom Dexter]

On Sat, May 1, 2010 at 1:44 PM, Tom Dexter <digitalaudiorock at gmail.com> wrote:
> On Wed, Apr 28, 2010 at 10:24 AM, John Drescher <drescherjm at gmail.com> wrote:
>>> Thanks John!  I might just try that.  I was going to say that you'd be
>>> pretty hard pressed to find any Gentoo user that configures the entire
>>> kernel from the defaults in order to upgrade.  If that's how you had
>>> to do it even I might quit Gentoo :D...and I'm a huge fanboy.  It's
>>> not like the kernel make process is very likely to let an invalid
>>> config get through or something.
>>>
>>> There's definitely something different in 2.6.32.  All those IOBOUND
>>> errors do in fact go away when I reboot to 2.6.31.
>>>
>>
>> I do not have 2.6.32 running on any of my mythtv boxes or any other
>> machine that I can think of at the moment. However I do run 2.6.33 on
>> several gentoo raid servers at work with ext4. I am not seeing any
>> IOBOUND errors there.
>>
>> John
>
> Well, I think I've figured a few things out:
>
> I installed the gentoo-sources 2.6.33-r1 kernel.  I then tested the
> system pretty maxed out: recording three DTV shows, commercial
> flagging one, and watching another recording from the frontend.  There
> were less IOBOUND errors, though I was still getting some.  They
> appeared to be primarily when recordings were starting, especially
> when several were starting at once.  Still not behaving as nicely as
> 2.6.31 though.
>
> However...I made an interesting discovery:  My EXT4 LVM has been at
> like 99% full for some time and has been auto-expiring old stuff I
> never got around to watching.  I deleted enough recordings to free up
> about 20GB of space...enough to go from 99% to about 96%.  After doing
> that I can't force any IOBOUND errors no matter how I max it out....at
> least not without trying crazy stuff like manually cp'ing huge files
> on the recording partition...which I'd expect to cause problems.
>
> I'm still unclear why this is affected by the kernel upgrade.  Maybe
> EXT4 in the newer kernels is somehow less forgiving of nearly full
> file systems(??).
>
> I'll keep an eye on it and see how it goes, but that seems to be the cause.
>
> Tom
>

This is interesting...I'm finding a lot of stuff about a change to
EXT4 in 2.6.32 that does in fact introduce a performance regression:

http://www.phoronix.com/scan.php?page=article&item=linux_perf_regressions&num=2

Interesting.  If I start getting the IOBOUND errors again once the
drive is more full I may try adding that nobarrier mount option and
see if they go away.  My EXT4 partition is only used for recordings
(nothing where data integrity is critical like the DB) where loosing a
cached write in a power outage is no issue...hell...the whole
recording is worthless at that point.

Tom