Bug#675934: st - slow tape read/write rate

[Jonathan Nieder]

Yury O. Tabolin wrote:

> Hi, so, I tried the kernels. First I tried to upgrade kernel to
> 2.6.32-45. Nothing has changed, the rate is same low.

Good, thanks for confirming.

[...]
> Then I installed 3.2.18-1~bpo60+1 kernel from squeeze-backports.
> Segfault disappeared. The rate increased slightly, but still

Ok, so now there are two different tasks to work on:

i) applying the improvement that eliminated the segfault to
squeeze

ii) fixing the rest of the speed regression relative to etch for
future kernels.

For (i): could you try 2.6.37-2 and 2.6.38-rc6 from
snapshot.debian.org and let us know how it goes?

For (ii): if you get a change to try 3.4.y from experimental, that
would be very useful. The only packages from outside squeeze aside
from the kernel itself that should be needed to test it are linux-base
and initramfs-tools, both of which can be found in squeeze-backports.

If the speed regression is still present, we can get help from
upstream. If it's fixed, we can try to find the patch that fixes it
and apply it to the wheezy and squeeze kernels. So either result is
progress.

Thanks again and hope that helps,
Jonathan



--
To UNSUBSCRIBE, email to debian-bugs-...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listm...@lists.debian.org

[Jonathan Nieder]

Yury O. Tabolin wrote:

> 20.06.2012 07:22, Jonathan Nieder wrote:

>> For (i): could you try 2.6.37-2 and 2.6.38-rc6 from
>> snapshot.debian.org and let us know how it goes?
>

> The result of trying 2.6.37-2 is same as 3.2 early. Segfault
> disappeared. The read rate is about 10-13 MB/sec.

Drat, ok. How about 2.6.36-1~experimental.1 or something near
2.6.33?

[...]

>> For (ii): if you get a change to try 3.4.y from experimental, that
>> would be very useful.
>

> With kernel 3.4.1-1 from exprimental nothing changed. No segfault.
> The rate is a some unstable, about 9-16 MB/sec.

Thanks much for checking. Please report the speed regression upstream
to linux...@vger.kernel.org, cc-ing Kai Mäkisara
<Kai.Ma...@kolumbus.fi> and either me or this bug log so we can
track it. Be sure to mention:

- steps to reproduce, expected result, actual result, and how the
difference indicates a bug (should be simple enough)

- which kernels you have tested and how each performed, roughly

- full "dmesg" output from booting and testing on an affected kernel,
as an attachment or link

- a link to this bug log for background

If we're lucky, someone might have ideas for tests to try to pin down
what in particular got worse. Or they might just ask you to bisect. :)

Thanks for making sure the st driver continues to work well, and sorry
I have no better ideas.

Hope that helps,

[Yury O. Tabolin]

27.06.2012 05:25, Jonathan Nieder пишет:

> Drat, ok. How about 2.6.36-1~experimental.1 or something near 2.6.33?
> [...]

So, sometimes later, I can to continue the experiments. I tried a few
kernels.

2.6.36-1~experimental.1 Segfault is absent, speed is slow.
# dd if=/dev/nst0 of=file bs=32k count=1500
1500+0 записей считано
1500+0 записей написано
скопировано 49152000 байт (49 MB), 3,87631 c, 12,7 MB/c
# modprobe -r st
# modprobe st
# dd if=/dev/nst0 of=file bs=32k count=1500
1500+0 записей считано
1500+0 записей написано
скопировано 49152000 байт (49 MB), 3,86245 c, 12,7 MB/c

2.6.34-1~experimental.2 Segfault is absent, speed is slow.
# dd if=/dev/nst0 of=file bs=32k count=1500
1500+0 записей считано
1500+0 записей написано
скопировано 49152000 байт (49 MB), 4,02346 c, 12,2 MB/c
# modprobe -r st
# modprobe st
# dd if=/dev/nst0 of=file bs=32k count=1500
1500+0 записей считано
1500+0 записей написано
скопировано 49152000 байт (49 MB), 3,52277 c, 14,0 MB/c

2.6.30-8squeeze1 Same as above, segfault is absent, speed is slow.
# dd if=/dev/nst0 of=file bs=32K count=1500
1500+0 записей считано
1500+0 записей написано
скопировано 49152000 байт (49 MB), 4,05358 c, 12,1 MB/c
# modprobe -r st
# modprobe st
# dd if=/dev/nst0 of=file bs=32k count=1500
1500+0 записей считано
1500+0 записей написано
скопировано 49152000 байт (49 MB), 4,01436 c, 12,2 MB/c

2.6.28-1 works perfectly. The speed is high, the same as before the
upgrade. This kernel I left work.
# dd if=/dev/nst0 of=file bs=32k count=1500
1500+0 записей считано
1500+0 записей написано
скопировано 49152000 байт (49 MB), 0,600784 c, 81,8 MB/c
# modprobe -r st
# modprobe st
# dd if=/dev/nst0 of=file bs=32k count=1500
1500+0 записей считано
1500+0 записей написано
скопировано 49152000 байт (49 MB), 0,559322 c, 87,9 MB/c
# dd if=/dev/nst0 of=file bs=32k count=5000
5000+0 записей считано
5000+0 записей написано
скопировано 163840000 байт (164 MB), 1,91439 c, 85,6 MB/c

> Please report the speed regression upstream
> to linux...@vger.kernel.org, cc-ing Kai Mäkisara
> <Kai.Ma...@kolumbus.fi> and either me or this bug log so we can
> track it.

Ok, I just did it. I hope this helps.

WBR, Yury O. Tabolin