[Bug 15884] New: cd-burning reduces other simultaneous IO performance

[Bug 15884] New: cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884

           Summary: cd-burning reduces other simultaneous IO performance
           Product: IO/Storage
           Version: 2.5
          Platform: All
        OS/Version: Linux
              Tree: Mainline
            Status: NEW
          Severity: normal
          Priority: P1
         Component: IDE
        AssignedTo: io_ide@kernel-bugs.osdl.org
        ReportedBy: danielhollocher@gmail.com
                CC: danielhollocher@gmail.com
        Regression: No


Hey folks, I having a problem similar to one of the commenters on bug #13347. 
But that report in general is kinda confusing, and it seemed like the devs had
trouble understanding the issue.  One possibility is that the issue that I and
the commenter have may not be the same as the original bug in 13347.  I also
have a reliable way to trigger the bug in a bad way, just burn a cd.  So, I am
posting this new bug report.

Basically, my system feels a bit sluggish, and when I try to burn a CD, my
system becomes unusable.  I will explain in further detail after I describe my
setup.

My system:
ASUS P5B motherboard
a 500GB SATA drive connect through one of the SATA ports (sda)
an 60 GB PATA drive, connect as the master on the single Jmicron IDE (sdb)
a Verbatim CD-RW, connected on the slave of the same IDE

My OSs:
I have windows on the first partition of sdb
I have my main Ubuntu install on another partition of sdb (linux version
2.6.31)
I have a testing Ubuntu install on one of the partitions of sda (linux version
2.6.32)

Now, I have always noticed that my main felt a bit sluggish.  Windows felt
faster, and I would just chock it up to "weirdness".  I suspect that I'm
getting poor performance in general, but it really shows up when I'm burning a
disk.  My computer becomes unusable, as any IO operations no matter how small
end up having several second wait times.  I have to be careful just moving a
window around.  I tested on windows, and I can burn a disk, defrag the drive,
and open open office all at the same time.  The system is correspondingly
sluggish, but that is to be expected.  I get no lockups like I do on linux.  I
have had this issue for the life of the computer, accross several ubuntu
releases.  I have also tester a kernel from the ubuntu mainline ppa of version
2.6.33.  

As you can see, I haven't tried downloading the source code and compiling
myself yet.  I don't know if that makes a difference.

Here are some illuminating tests.  I was booted on the nice drive (sda) with
the testing install.

$ sudo hdparm -tT /dev/sdb
/dev/sdb:
 Timing cached reads:   2146 MB in  2.00 seconds = 1073.83 MB/sec
 Timing buffered disk reads:  128 MB in  3.03 seconds =  42.27 MB/sec
$ sudo hdparm -tT /dev/sda
/dev/sda:
 Timing cached reads:   2308 MB in  2.00 seconds = 1154.93 MB/sec
 Timing buffered disk reads:  284 MB in  3.01 seconds =  94.50 MB/sec


I started the burn process, and caught this test while it was erasing the CDRW:
/dev/sdb:
 Timing cached reads:   2294 MB in  2.00 seconds = 1147.90 MB/sec
 Timing buffered disk reads:    2 MB in 88.41 seconds =  23.16 kB/sec


During the burn process, which displays a speed of 0.7MB/s:
$ sudo hdparm -tT /dev/sdb
/dev/sdb:
 Timing cached reads:   908 MB in  2.00 seconds = 453.90 MB/sec
 Timing buffered disk reads:    6 MB in  3.87 seconds =   1.55 MB/sec
$ sudo hdparm -tT /dev/sda
/dev/sda:
 Timing cached reads:   2254 MB in  2.00 seconds = 1127.28 MB/sec
 Timing buffered disk reads:  282 MB in  3.00 seconds =  93.98 MB/sec


So while burning on the slave of the IDE, I loose almost all my "buffered disk
read" performance for the master IDE channel!  Does that make sense?  If I burn
a cd at 1MB/s, it should not take up 41MB/s bandwidth.

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.

[Bug 15884] cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884





--- Comment #1 from Dan Hollocher <danielhollocher@gmail.com>  2010-04-30 15:48:45 ---
I originally filed a bug on launchpad here:
https://bugs.launchpad.net/linux/+bug/262867

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.

[Bug 15884] cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884


Robert Hancock <hancockrwd@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hancockrwd@gmail.com




--- Comment #2 from Robert Hancock <hancockrwd@gmail.com>  2010-05-01 01:29:37 ---
You can't have a command outstanding on both the master and slave drive on a
PATA channel at the same time. If the software issues a command like "erase
disc" to the drive, during the time that command is outstanding no commands can
be issued to the other drive on that channel, it's the way PATA works.

Now, there may be a difference in what commands the burning software is using
or the ordering that results in some of them taking longer than it does in
Windows or something. (I don't know the MMC commands that well but I think some
of them do have a mode where you can issue the command and poll periodically to
see if it's done rather than blocking on the operation until it finishes.) That
isn't anything the kernel can do anything about, though - if something issues a
command to one drive on the channel that takes a long time, anything wanting to
access the other drive just has to wait.

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.

Re: [Bug 15884] cd-burning reduces other simultaneous IO performance

[Stan Hoeppner]

bugzilla-daemon@bugzilla.kernel.org put forth on 4/30/2010 8:29 PM:

> Now, there may be a difference in what commands the burning software is using
> or the ordering that results in some of them taking longer than it does in
> Windows or something. (I don't know the MMC commands that well but I think some
> of them do have a mode where you can issue the command and poll periodically to
> see if it's done rather than blocking on the operation until it finishes.) That
> isn't anything the kernel can do anything about, though - if something issues a
> command to one drive on the channel that takes a long time, anything wanting to
> access the other drive just has to wait.

This has been known for over a decade.  When I was building white box
Windows systems in the mid-late 90s our SOP was to put burners on a
dedicated IDE port specifically due to what the OP reports over a decade
later as a bug.  As Robert points out above, this isn't a software bug, but
a hardware limitation of IDE.  Parallel SCSI, SAS, and SATA do not suffer
this limitation.  Apparently Windows engineers have developed a method over
the years to mitigate the ill effects of this.  I can tell you from
experience that Windows 9x/ME/NT4/W2K all suffered performance degradation
with HD+burner on the same channel problem.

-- 
Stan

[Bug 15884] cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884


Tejun Heo <tj@kernel.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tj@kernel.org




--- Comment #3 from Tejun Heo <tj@kernel.org>  2010-05-05 09:46:20 ---
Hmm... can you please attach kernel boot log?  They don't need to occupy the
same channel.  Maybe windows is switching the controller to another mode?

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.

[Bug 15884] cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884





--- Comment #4 from Dan Hollocher <danielhollocher@gmail.com>  2010-05-05 14:55:42 ---
Created an attachment (id=26235)
 --> (https://bugzilla.kernel.org/attachment.cgi?id=26235)
/var/log/kern.log from the sda/testing install, after I burned a cd

the I/O errors on sr0 happen when I put a disk in the drive  (I've seen this
with 3 RW disks.  I don't have any other non-RW disks).  The kjournald thing
happened when I started k3b.  I'm pretty sure the later I/O errors happened
when the disk burning was completed.

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.

[Bug 15884] cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884





--- Comment #5 from Robert Hancock <hancockrwd@gmail.com>  2010-05-05 23:22:48 ---
That part of the log isn't very interesting, obviously someone's trying to read
from the disc and the drive's saying no (possibly because the disc's still
blank). Can you post the output of dmesg after bootup?

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.

[Bug 15884] cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884


Dan Hollocher <danielhollocher@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #26235|0                           |1
        is obsolete|                            |




--- Comment #6 from Dan Hollocher <danielhollocher@gmail.com>  2010-05-06 01:08:32 ---
Created an attachment (id=26252)
 --> (https://bugzilla.kernel.org/attachment.cgi?id=26252)
fresh boot dmesg

oops, I'm sorry.  I didn't realize that kern.log would contain so little.

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.

[Bug 15884] cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884





--- Comment #7 from Robert Hancock <hancockrwd@gmail.com>  2010-05-06 01:49:17 ---
Ah, it's one of those JMicron PATA controllers. I'm assuming you only have one
physical PATA connector on the board, so splitting the devices onto different
channels isn't an option.

Unfortunately I don't think there's much that can be done about this at the
kernel level, it's kind of an inherent limitation of PATA. The burning software
might be able to improve on things with different selection/ordering of
commands, but the kernel doesn't have any control over that, it just passes the
commands through to the drive.

Honestly I think the best solution here is to either add another PATA
controller to provide another channel, or replace one of the devices with a
SATA one.

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.

[Bug 15884] cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884





--- Comment #8 from Robert Hancock <hancockrwd@gmail.com>  2010-05-13 02:24:17 ---
The -immed flag to wodim may improve the situation, if you can make k3b add it
to the wodim command line somehow:

       -immed Tell  wodim  to  set  the  SCSI  IMMED  flag in certain commands
              (load/eject/blank/close_track/close_session).  This can be  use-
              ful  on  broken systems with ATAPI harddisk and CD/DVD writer on
              the same bus  or  with  SCSI  systems  that  don’t  use  discon-
              nect/reconnect.   These  systems  will  freeze while blanking or
              fixating a CD/DVD or while a DVD writer is filling up a  session
              to the minimum amount (approx. 800 MB).  Setting the -immed flag
              will request the command to return immediately while the  opera-
              tion proceeds in background, making the bus usable for the other
              devices and avoiding the system freeze.  This is an experimental
              feature  which  may  work  or not, depending on the model of the
              CD/DVD writer.  A correct solution would be to set up a  correct
              cabling but there seem to be notebooks around that have been set
              up the wrong way by the manufacturer.  As it  is  impossible  to
              fix this problem in notebooks, the -immed option has been added.

              A second experimental feature of the  -immed  flag  is  to  tell
              wodim  to  try  to  wait short times while writing to the media.
              This is expected to free the IDE bus if the  CD/DVD  writer  and
              the  data  source  are  connected to the same IDE cable. In this
              case, the CD/DVD writer would otherwise usually  block  the  IDE
              bus  for  nearly all the time making it impossible to fetch data
              from the source drive. See also minbuf= and -v option.

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.

[Bug 15884] cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884





--- Comment #9 from Dan Hollocher <danielhollocher@gmail.com>  2010-05-13 20:57:19 ---
That is an awesome find Robert!  I will test and report back.  Btw, I tested
with nerolinux, and that performs allot better, so it looks like you are
correct.  Testing with -immed should firm that up.

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.

[Bug 15884] cd-burning reduces other simultaneous IO performance

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=15884


Dan Hollocher <danielhollocher@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |INVALID




--- Comment #10 from Dan Hollocher <danielhollocher@gmail.com>  2010-05-14 01:20:19 ---
Tested and confirmed.

Here are directions on how to apply this to k3b (for others that may stop by):
Settings > Configure K3b... > Programs > User Parameters
Click near cdrecord, and add the option -immed.

You may need to ensure that cdrecord is selected by:
Settings > Configure K3b... > Advanced > Show advanced GUI elements
During the burning setup, you will have an option to select cdrecord.

-- 
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are watching the assignee of the bug.