2003-07-22 21:27:51

by Mike Fedyk

[permalink] [raw]
Subject: different behaviour with badblocks on 2.6.0-test1-mm1-07int

Hi,

I was testing a hard drive with badblocks (from the e2fsprogs-1.34) on the
2.6.0-test1-mm1-07int (with Con's scheduler patch), and I noticed in vmstat
and gkrellm that during the write passes there are reads on the same drive
when there should only be writes.

I tried stracing badblocks, but all it showed was write() calls, and vmstat
and gkrellm showed reads only, so it modified the behaviour.

Has anyone else seen this?

ii e2fsprogs 1.33+1.34-WIP-2003.05 The EXT2 file system
utilities and libraries


2003-07-22 21:30:56

by Mike Fedyk

[permalink] [raw]
Subject: Re: different behaviour with badblocks on 2.6.0-test1-mm1-07int

On Tue, Jul 22, 2003 at 02:42:53PM -0700, Mike Fedyk wrote:
> Hi,
>
> I was testing a hard drive with badblocks (from the e2fsprogs-1.34) on the
> 2.6.0-test1-mm1-07int (with Con's scheduler patch), and I noticed in vmstat
> and gkrellm that during the write passes there are reads on the same drive
> when there should only be writes.
>
> I tried stracing badblocks, but all it showed was write() calls, and vmstat
> and gkrellm showed reads only, so it modified the behaviour.
>
> Has anyone else seen this?
>
> ii e2fsprogs 1.33+1.34-WIP-2003.05 The EXT2 file system
> utilities and libraries
>

Oh, and testing with the same hardware and userspace on 2.4.22-pre7 shows
normal behaviour (writes with no reading, reads with no writing).

This is with "badblocks -wso /tmp/hde.out /dev/hde > /dev/hde.log 2>&1 &" on
a bash prompt on both kernels.

Neither found any bad blocks, and /tmp is on a /dev/hda1

2003-07-22 22:00:30

by Mike Fedyk

[permalink] [raw]
Subject: Re: different behaviour with badblocks on 2.6.0-test1-mm1-07int

On Tue, Jul 22, 2003 at 02:45:57PM -0700, Mike Fedyk wrote:
> On Tue, Jul 22, 2003 at 02:42:53PM -0700, Mike Fedyk wrote:
> > Hi,
> >
> > I was testing a hard drive with badblocks (from the e2fsprogs-1.34) on the
> > 2.6.0-test1-mm1-07int (with Con's scheduler patch), and I noticed in vmstat
> > and gkrellm that during the write passes there are reads on the same drive
> > when there should only be writes.
> >
> > I tried stracing badblocks, but all it showed was write() calls, and vmstat
> > and gkrellm showed reads only, so it modified the behaviour.
> >
> > Has anyone else seen this?
> >
> > ii e2fsprogs 1.33+1.34-WIP-2003.05 The EXT2 file system
> > utilities and libraries
> >
>
> Oh, and testing with the same hardware and userspace on 2.4.22-pre7 shows
> normal behaviour (writes with no reading, reads with no writing).
>
> This is with "badblocks -wso /tmp/hde.out /dev/hde > /dev/hde.log 2>&1 &" on
> a bash prompt on both kernels.
>
> Neither found any bad blocks, and /tmp is on a /dev/hda1
>

This is also being reported against e2fsprogs in debian to get more eyes on
it.

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=202008

2003-07-23 08:04:07

by Apurva Mehta

[permalink] [raw]
Subject: Re: different behaviour with badblocks on 2.6.0-test1-mm1-07int

* Mike Fedyk <[email protected]> [2003-07-23 12:01]:
> Hi,
>
> I was testing a hard drive with badblocks (from the e2fsprogs-1.34) on the
> 2.6.0-test1-mm1-07int (with Con's scheduler patch), and I noticed in vmstat
> and gkrellm that during the write passes there are reads on the same drive
> when there should only be writes.
>
> I tried stracing badblocks, but all it showed was write() calls, and vmstat
> and gkrellm showed reads only, so it modified the behaviour.
>
> Has anyone else seen this?

On 2.6.0-test1, gkrellm does not show any disk usage at all for
me. The disk usage krell just remains blank. vmstat reports expected
usage though.

- Apurva

2003-07-23 13:09:12

by Valdis Klētnieks

[permalink] [raw]
Subject: Re: different behaviour with badblocks on 2.6.0-test1-mm1-07int

On Wed, 23 Jul 2003 13:48:44 +0530, Apurva Mehta <[email protected]> said:

> On 2.6.0-test1, gkrellm does not show any disk usage at all for
> me. The disk usage krell just remains blank. vmstat reports expected
> usage though.

Upgrade your gkrellm - 2.1.14 is current.

2.1.6 Wed Jan 22, 2003
...
* Patches:
o Andreas Boman <aboman--at--eiwaz.com> had two Linux patches:
...
- implemented reading disk stats from sysfs for recent 2.5.x kernels.


Attachments:
(No filename) (226.00 B)

2003-07-23 15:05:14

by Apurva Mehta

[permalink] [raw]
Subject: Re: different behaviour with badblocks on 2.6.0-test1-mm1-07int

* [email protected] <[email protected]> [2003-07-23 19:55]:
> On Wed, 23 Jul 2003 13:48:44 +0530, Apurva Mehta <[email protected]> said:
>
> > On 2.6.0-test1, gkrellm does not show any disk usage at all for
> > me. The disk usage krell just remains blank. vmstat reports expected
> > usage though.
>
> Upgrade your gkrellm - 2.1.14 is current.
>
> 2.1.6 Wed Jan 22, 2003
> ...
> * Patches:
> o Andreas Boman <aboman--at--eiwaz.com> had two Linux patches:
> ...
> - implemented reading disk stats from sysfs for recent 2.5.x kernels.

I am using 2.1.14.. Still no luck with it reading disks..

- Apurva

2003-07-23 16:46:03

by Mike Fedyk

[permalink] [raw]
Subject: Re: different behaviour with badblocks on 2.6.0-test1-mm1-07int

On Wed, Jul 23, 2003 at 08:49:50PM +0530, Apurva Mehta wrote:
> * [email protected] <[email protected]> [2003-07-23 19:55]:
> > On Wed, 23 Jul 2003 13:48:44 +0530, Apurva Mehta <[email protected]> said:
> >
> > > On 2.6.0-test1, gkrellm does not show any disk usage at all for
> > > me. The disk usage krell just remains blank. vmstat reports expected
> > > usage though.
> >
> > Upgrade your gkrellm - 2.1.14 is current.
> >
> > 2.1.6 Wed Jan 22, 2003
> > ...
> > * Patches:
> > o Andreas Boman <aboman--at--eiwaz.com> had two Linux patches:
> > ...
> > - implemented reading disk stats from sysfs for recent 2.5.x kernels.
>
> I am using 2.1.14.. Still no luck with it reading disks..

Maybe you need to mount sysfs on /sys?

2003-07-23 18:54:50

by Apurva Mehta

[permalink] [raw]
Subject: Re: different behaviour with badblocks on 2.6.0-test1-mm1-07int

* Mike Fedyk <[email protected]> [2003-07-23 22:41]:
> > I am using 2.1.14.. Still no luck with it reading disks..
>
> Maybe you need to mount sysfs on /sys?

Right, gkrellm does report some disk usage now, but it is far from
accurate. It registers barely any of the disk activity. Starting
Firebird or Opera may occasionally register one spike on the
graph. Mostly, the disk activity is not reported.

This seems to be a gkrellm bug since vmstat reports disk usage more
accurately, although I haven't really looked closely at the output..

- Apurva

2003-07-23 19:21:31

by Mike Fedyk

[permalink] [raw]
Subject: Re: different behaviour with badblocks on 2.6.0-test1-mm1-07int

On Thu, Jul 24, 2003 at 12:39:29AM +0530, Apurva Mehta wrote:
> * Mike Fedyk <[email protected]> [2003-07-23 22:41]:
> > > I am using 2.1.14.. Still no luck with it reading disks..
> >
> > Maybe you need to mount sysfs on /sys?
>
> Right, gkrellm does report some disk usage now, but it is far from
> accurate. It registers barely any of the disk activity. Starting
> Firebird or Opera may occasionally register one spike on the
> graph. Mostly, the disk activity is not reported.
>
> This seems to be a gkrellm bug since vmstat reports disk usage more
> accurately, although I haven't really looked closely at the output..

Ok let's forget about gkrellm because I see similar numbers in vmstat.

Now, let's get back to the origional issue, which is "Why the fuck is
badblocks reading when it should only be writing, and why does it only
happen on a 2.6-test kernel?!!"