please enable CONFIG_TASKSTATS, especially CONFIG_TASK_IO_ACCOUNTING

Bug #157191 reported by Pasi Savolainen
20
Affects Status Importance Assigned to Milestone
linux (Ubuntu)
Fix Released
Low
Tim Gardner

Bug Description

Linux kernel has very useful task IO information (reflected into /proc/self/io and other places) similar to what 'vmstat' gives.
It needs that in kernel configuration CONFIG_TASKSTATS and CONFIG_TASK_IO_ACCOUNTING are enabled.
In current gutsy (2.6.22.14.21) kernel they are not enabled.

There is also very handy 'iotop.py' program available at http://guichaz.free.fr/misc/

Revision history for this message
gmayer (gunther-mayer) wrote :

I need to second this request. Being able to view disk i/o stats per process is very important as it would greatly help newbies, advanced users and developers alike to track down what i/o hogging processes are causing their ubuntu systems to slow down. This happens more than people think, the recent spate of bugs/forum posts related to trackerd is just one example of how top/vmstat/iostat et al are useless for such a task.

People are asking for this capability all the time, see

http://ubuntuforums.org/showthread.php?t=646611
http://ubuntuforums.org/showthread.php?t=294555
http://ubuntuforums.org/showthread.php?t=396692

as just some examples of users screaming for this functionality. While I'm an avid Ubuntu user myself (I completely wiped my M$ Windoze partition btw) I think it's pretty sad that all other OS's have that capability built-in but ubuntu doesn't, and all it takes to fix is to turn on a simple kernel config option...

BTW, I got alerted to this by an excellent thread on linuxquestions

http://www.linuxquestions.org/questions/linux-kernel-70/how-to-enable-process-io-statistics-600777/#post2963872

which mentions that the latest version of collectl also supporting the "new" i/o /proc interface (not in the repos yet, see bug# 135037)

Revision history for this message
Nicolò Chieffo (yelo3) wrote :

yes, please! I also think it is a great option!

Revision history for this message
Jo-Erlend Schinstad (joerlend.schinstad-deactivatedaccount) wrote :

Yes, this is important. The question is if there is any reason for it not to be configured that way? Such functionality becomes even more important as virtualization becomes more common, I think. I/O is probably the biggest bottleneck there is atm, so you need to track down the source as quickly as possible. Having iotop available could greatly simplify this.

Changed in linux:
assignee: nobody → timg-tpi
importance: Undecided → Low
milestone: none → hardy-alpha-4
status: New → In Progress
Revision history for this message
Nicolò Chieffo (yelo3) wrote :

2.6.24-4-generic has this feature enabled by default (just tested iotop.py and it is running)

Changed in linux:
status: In Progress → Fix Released
Revision history for this message
dahias (wengahias) wrote :

please turn it on again ! it would be very nice to have this feature for our Serverfarm enabled because there is no good alternative to analyze and improve performance of the daily I/O Traffic caused by virtual guests.

Revision history for this message
Nicolò Chieffo (yelo3) wrote : Re: [Bug 157191] Re: please enable CONFIG_TASKSTATS, especially CONFIG_TASK_IO_ACCOUNTING

Isn't it turned on? it works with me

Revision history for this message
gmayer (gunther-mayer) wrote :

It's working for me in hardy, my original post was still for gutsy. Thanks guys, it's VERY useful to see per process i/o stats now...

I think this bug should be closed now.

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.