lkml.org 
[lkml]   [2011]   [Jul]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: App blocked in futex() burns 14% CPU.


On 07/14/2011 03:51 PM, Chris Friesen wrote:
> On 07/14/2011 04:34 PM, Jesper Juhl wrote:
>> So, I've recently started enabling the KDE desktop search on my box and it
>> runs some indexing services (naturally) to keep track of changes to files.
>> One of the apps it starts is a process named "virtuoso-t". This process
>> burns a *lot* of CPU when it's just started, but once it has caught up
>> with recent changes it quiets down. It doesn't quite quiet down to the
>> level I'd expect though. I see it constantly using 12-14% CPU time in
>> 'top' even when there is *nothing* going on on the machine :
>>
>> top - 00:25:09 up 2:38, 2 users, load average: 0.01, 0.04, 0.05
>> Tasks: 155 total, 1 running, 153 sleeping, 0 stopped, 1 zombie
>> Cpu(s): 0.7%us, 0.8%sy, 3.7%ni, 94.6%id, 0.1%wa, 0.0%hi, 0.0%si, 0.0%st
>> Mem: 3853028k total, 2154696k used, 1698332k free, 184280k buffers
>> Swap: 4200992k total, 0k used, 4200992k free, 996824k cached
>>
>> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
>> 1952 jj 39 19 362m 263m 6544 S 14 7.0 25:09.54 virtuoso-t
>> 1811 jj 20 0 635m 27m 17m S 2 0.7 2:49.78 knotify4
>> 1928 jj 39 19 595m 26m 18m S 1 0.7 2:58.40 nepomukservices
>>
>> This box has a dual core Intel core i5-560M CPU with hyperthreading, so it
>> is 12-14% of a fairly capable CPU - that's rather a lot more than I'd
>> expect when the box is idle and there's nothing for the process to do.
>>
>> So, I tried strace'ing the process and it seems to just be blocked in
>> futex() most of the time (like 99%) :
>>
>> [jj@dragon ~]$ strace -p 1952
>> Process 1952 attached - interrupt to quit
>> futex(0x3086424, FUTEX_WAIT_PRIVATE, 503, NULL

Try strace with -f. If you really want to collect statistics about where
it is spending it's time, consider oprofile or perf.

>>
>> So it's just sitting there doing nothing, which lead me to think that this
>> is probably not a bug in the application - if it was doing some silly
>> polling loop or similar I would not just see it blocked in a futex() call.
>> So I'm guessing something must be happening in kernel space that's burning
>> a lot of CPU.. I could be completely wrong of course, and if I am, please
>> feel free to enlighten me.
>
> You don't have enough kernel activity to account for 14% of a cpu. Have
> you checked whether there's another thread within the process that's
> doing something in userspace?
>

Agreed.

--
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel


\
 
 \ /
  Last update: 2011-07-15 20:55    [W:0.065 / U:2.796 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site