[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CA+icZUVbONNq2zuUqo1_OZTaNrDCUiWBGURKtv0MAqF=GoS8Mw@mail.gmail.com>
Date: Sat, 12 Jan 2013 13:54:36 +0100
From: Sedat Dilek <sedat.dilek@...il.com>
To: Dave Kleikamp <dave.kleikamp@...cle.com>
Cc: linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org,
Zach Brown <zab@...bo.net>,
"Maxim V. Patlasov" <mpatlasov@...allels.com>
Subject: Re: [PATCH V5 00/30] loop: Issue O_DIRECT aio using bio_vec
On Sat, Jan 12, 2013 at 11:52 AM, Sedat Dilek <sedat.dilek@...il.com> wrote:
> On Fri, Jan 11, 2013 at 10:51 PM, Dave Kleikamp
> <dave.kleikamp@...cle.com> wrote:
>> On 01/10/2013 09:46 AM, Sedat Dilek wrote:
>>
>>> Hi Dave,
>>>
>>> I am using here Ubuntu/precise AMD64 as a WUBI-installed system.
>>>
>>> Not sure if WUBI [1] is a good test-candidate.
>>>
>>> [ /boot/grub/grub.cfg ]
>>> ...
>>> set root=(loop0)
>>> linux /boot/vmlinuz-3.8.0-rc2-9-iniza-generic
>>> root=UUID=001AADA61AAD9964 loop=/ubuntu/disks/root.disk ro
>>> ...
>>>
>>> Poops, I did it (not again) but for the first time to test your
>>> loop-patchset on top of Linux v3.8-rc2 plus some important other stuff
>>> (see patches/ dir in attached tarball).
>>>
>>> As I did not know how to test it in a meaningful way I just run the
>>> "lite" test-script from LTP [2].
>>> Please, have a look at the ERRORs and failures.
>>
>> I'd really like to know if these same testcases fail without my
>> patchset. I'm going to play with ltp on loop-mounted filesystems, but I
>> don't intend to play with WUBI at all.
>>
>
> I am preparing now two kernels against Linux v3.8-rc3 (as your
> patchset applies cleany) with 2 important patches which I need here
> (1. mei: no proper reboot and 2. (S)ATA fix).
>
> I will attach two tarballs for each kernel-setup and run LTP "lite"
> again (tarballs will include the results).
>
As promised both tarballs attached!
*-iniza-generic: Linux v3.8-rc3 with my patches from patches/ dir
*-shaggy-generic: Identical as above, but with your loop patchset (see
patch in patches/ dir).
$ egrep -i 'error|fail'
v3.8.0-rc3-4-iniza-generic/tests/runltplite-results_*.txt | egrep -v
-i 'expected|perm' | wc -l
197
$ egrep -i 'error|fail'
v3.8.0-rc3-5-shaggy-generic/tests/runltplite-results_*.txt | egrep -v
-i 'expected|perm' | wc -l
196
As said I am new to LTP and might have a deeper look into the results files.
( I am really interested in a loop-benchmark, but have no real good idea. )
Hope this helps you!
- Sedat -
>> I'd be interested if my patchset introduces regressions, but not so much
>> if the same testcases fail previously.
>>
>
> Hmm, regressions is always good to test.
> As a "customer" aka tester I want to see any benefit means for most
> power-users: Do I get some speedups?
> That's the main background of my askings.
>
> As said here I am running Ubuntu as a WUBI-installation.
> This system is predestinated for testing loop improvements.
> So again and no sorry: How can I test reliable speed improvements?
> I remember linux-fs/linux-xfs folks have a tool could be named
> "xfstests" (note2myself: more coffee!).
> Any hints for testing appreciated!
>
> - Sedat -
>
>>> $ egrep -i 'error|fail'
>>> for-dkleikamp/tests/runltplite-results_loop-experimental.txt | grep -v
>>> -i expected | wc -l
>>> 210
>>>
>>> In good old German tradition I have collected some interesting stuff
>>> in the attached tarball ;-).
>>> If something is missing - blame me.
>>> Don't hesitate to ask (I have your patchset for a while on my radar).
>>>
>>> Thanks!
>>>
>>> Regards,
>>> - Sedat -
>>>
>>> [1] https://wiki.ubuntu.com/WubiGuide
>>> [2] http://sourceforge.net/projects/ltp/
Download attachment "v3.8.0-rc3-4-iniza-generic.tar.xz" of type "application/octet-stream" (74804 bytes)
Download attachment "v3.8.0-rc3-4-iniza-generic.tar.xz.sha256sum" of type "application/octet-stream" (100 bytes)
Download attachment "v3.8.0-rc3-5-shaggy-generic.tar.xz" of type "application/octet-stream" (105296 bytes)
Download attachment "v3.8.0-rc3-5-shaggy-generic.tar.xz.sha256sum" of type "application/octet-stream" (101 bytes)
Powered by blists - more mailing lists