** Attachment added: "BootDmesg.txt"
   http://launchpadlibrarian.net/31580108/BootDmesg.txt

** Attachment added: "CurrentDmesg.txt"
   http://launchpadlibrarian.net/31580110/CurrentDmesg.txt

** Attachment added: "Dependencies.txt"
   http://launchpadlibrarian.net/31580112/Dependencies.txt

** Attachment added: "HalComputerInfo.txt"
   http://launchpadlibrarian.net/31580113/HalComputerInfo.txt

** Attachment added: "Lspci.txt"
   http://launchpadlibrarian.net/31580114/Lspci.txt

** Attachment added: "Lsusb.txt"
   http://launchpadlibrarian.net/31580115/Lsusb.txt

** Attachment added: "ProcCpuinfo.txt"
   http://launchpadlibrarian.net/31580116/ProcCpuinfo.txt

** Attachment added: "ProcInterrupts.txt"
   http://launchpadlibrarian.net/31580117/ProcInterrupts.txt

** Attachment added: "ProcModules.txt"
   http://launchpadlibrarian.net/31580118/ProcModules.txt

** Description changed:

  Ubuntu currently uses the kernel default i/o scheduler, CFQ. I strongly
  believe that this is the wrong scheduler to use for the vast majority of
  likely Ubuntu desktop installs because it causes huge lag in interactive
  applications under the heavy kind of i/o typically experienced during a
  file copy or backup operation.
  
  DISCLAIMER: Of course it's always hard to justify these sorts of claims
  because desktop responsiveness is a notoriously woolly and hard to
  measure thing, but without at least suggesting some solutions I don't
  see how we can improve it, hence this bug report.
  
  With that said, I would politely request that the Ubuntu kernel team at
  least examine the performance, and very importantly, the general desktop
  responsiveness experienced when using a variety of different i/o
  schedulers. I also think this could go a long way to solving bug no.
- 131094 (https://bugs.launchpad.net/ubuntu/+source/linux-
- source-2.6.22/+bug/131094).
+ 131094 .
  
  To reproduce on a standard ubuntu desktop, open (for example) a music
  player, firefox and/or any other slightly latency sensitive ui
  applications. Now execute:
  
  sudo dd if=/dev/$ROOTPARTITION of=/dev/zero bs=10M
  
  where $ROOTPARTITION is your root partition eg sda1, hdb2 or whatever.
  This should simulate the heavy i/o likely to be experienced whilst doing
  a large file copy/backup or similar. Now go back to normal web
  browsing/email writing/whatever and observe some truly horrible ui lag
  and general unresponsiveness.
  
  The current i/o scheduler for a particular partition can be changed on
  the fly by editing /sys/block/$PARTITION/queue/scheduler. "cat
  /sys/block/$PARTITION/queue/scheduler" will tell you the current i/o
  scheduler in use as well as other possible other schedulers that can be
  used. To change it, you must become root with sudo -s and then "echo
  $SCHEDULER > /sys/block/sdx/queue/scheduler". The i/o scheduler can be
  set to a different default globally by appending the elevator=$SCHEDULER
  option to the grub kernel boot stanza.
  
  Try changing the i/o scheduler for your root  partition to "deadline"
  for example and repeat the test. I observed MASSIVELY improved ui
  responsiveness during the heavy i/o test on my desktop, and several
  others who I have discussed this with have seen similar improvements.
  Now I'm certainly NOT saying "switch to the deadline i/o scheduler", but
  I do think the ubuntu kernel devs should experiment a bit and see if
  perhaps the current i/o scheduler is not the best option for desktop
  responsiveness. After all, a scheduler that copes well with the typical
  tasks a server has to deal with may not also cope as well with the tasks
  a desktop has to deal with and vice versa. For Ubuntu Desktop, perceived
  desktop performance and interactivity should come first, so perhaps the
  kernel defaults here (which could possibly be tuned more for server-type
  workloads) may not be the best option.
  
  ProblemType: Bug
  Architecture: amd64
  DistroRelease: Ubuntu 9.04
  MachineType: System manufacturer P5QL PRO
  NonfreeKernelModules: nvidia
  Package: linux-image-2.6.28-15-generic 2.6.28-15.49
  ProcCmdLine: root=UUID=f9c198f6-df39-49bf-9444-a1ad71d16bf6 ro 
elevator=deadline quiet splash
  ProcEnviron:
   LANG=en_GB.UTF-8
   SHELL=/bin/bash
  ProcVersionSignature: Ubuntu 2.6.28-15.49-generic
  SourcePackage: linux

** Description changed:

  Ubuntu currently uses the kernel default i/o scheduler, CFQ. I strongly
  believe that this is the wrong scheduler to use for the vast majority of
  likely Ubuntu desktop installs because it causes huge lag in interactive
  applications under the heavy kind of i/o typically experienced during a
  file copy or backup operation.
  
  DISCLAIMER: Of course it's always hard to justify these sorts of claims
  because desktop responsiveness is a notoriously woolly and hard to
  measure thing, but without at least suggesting some solutions I don't
  see how we can improve it, hence this bug report.
  
  With that said, I would politely request that the Ubuntu kernel team at
  least examine the performance, and very importantly, the general desktop
  responsiveness experienced when using a variety of different i/o
  schedulers. I also think this could go a long way to solving bug no.
- 131094 .
+ 131094 , especially since numerous users in that bug thread have
+ reported significant improvements in responsiveness after changing i/o
+ schedulers.
  
  To reproduce on a standard ubuntu desktop, open (for example) a music
  player, firefox and/or any other slightly latency sensitive ui
  applications. Now execute:
  
  sudo dd if=/dev/$ROOTPARTITION of=/dev/zero bs=10M
  
  where $ROOTPARTITION is your root partition eg sda1, hdb2 or whatever.
  This should simulate the heavy i/o likely to be experienced whilst doing
  a large file copy/backup or similar. Now go back to normal web
  browsing/email writing/whatever and observe some truly horrible ui lag
  and general unresponsiveness.
  
  The current i/o scheduler for a particular partition can be changed on
  the fly by editing /sys/block/$PARTITION/queue/scheduler. "cat
  /sys/block/$PARTITION/queue/scheduler" will tell you the current i/o
  scheduler in use as well as other possible other schedulers that can be
  used. To change it, you must become root with sudo -s and then "echo
  $SCHEDULER > /sys/block/sdx/queue/scheduler". The i/o scheduler can be
  set to a different default globally by appending the elevator=$SCHEDULER
  option to the grub kernel boot stanza.
  
  Try changing the i/o scheduler for your root  partition to "deadline"
  for example and repeat the test. I observed MASSIVELY improved ui
  responsiveness during the heavy i/o test on my desktop, and several
  others who I have discussed this with have seen similar improvements.
  Now I'm certainly NOT saying "switch to the deadline i/o scheduler", but
  I do think the ubuntu kernel devs should experiment a bit and see if
  perhaps the current i/o scheduler is not the best option for desktop
  responsiveness. After all, a scheduler that copes well with the typical
  tasks a server has to deal with may not also cope as well with the tasks
  a desktop has to deal with and vice versa. For Ubuntu Desktop, perceived
  desktop performance and interactivity should come first, so perhaps the
  kernel defaults here (which could possibly be tuned more for server-type
  workloads) may not be the best option.
  
  ProblemType: Bug
  Architecture: amd64
  DistroRelease: Ubuntu 9.04
  MachineType: System manufacturer P5QL PRO
  NonfreeKernelModules: nvidia
  Package: linux-image-2.6.28-15-generic 2.6.28-15.49
  ProcCmdLine: root=UUID=f9c198f6-df39-49bf-9444-a1ad71d16bf6 ro 
elevator=deadline quiet splash
  ProcEnviron:
   LANG=en_GB.UTF-8
   SHELL=/bin/bash
  ProcVersionSignature: Ubuntu 2.6.28-15.49-generic
  SourcePackage: linux

-- 
CFQ may not be the right choice of i/o scheduler for the most common desktop 
systems
https://bugs.launchpad.net/bugs/427210
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.

-- 
ubuntu-bugs mailing list
ubuntu-bugs@lists.ubuntu.com
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to