Re: wired top (and others) behavior - broken CPU usage reporting ?

Vladimir Grebenschikov wrote:
> Hi 
> 
> I have notice that it is no more possible to find what process is eats
> all CPU time with top, vmstat and tools like that. This happens with
> current some (big) time ago.
> 
> See here vmstat (vm mode) output:
> 
>  3.3%Sys   0.8%Intr 95.9%User  0.0%Nice  0.0%Idle        %ozfod   128
> rtc irq8
> |    |    |    |    |    |    |    |    |    |    |       daefr    66
> cbb0 pcm0+
> ==>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>        prcfr     1
> ata0 irq14
>                                         33 dtbuf       47 totfr
> ata1 irq15
> Namei     Name-cache   Dir-cache     69694 desvn          react
>    Calls    hits   %    hits   %     15144 numvn          pdwak
>     1259    1259 100                  9063 frevn          pdpgs
>                                                           intrn
> 
> 99.5% CPU used in user space.
> 
> now top output (sorted by CPU):
> last pid:  2024;  load averages:  1.03,  0.65, 0.40 up 0+01:28:34
> 11:38:50
> 120 processes: 4 running, 116 sleeping
> CPU states: 95.7% user,  0.0% nice,  3.5% system,  0.8% interrupt,  0.0%
> idle
> Mem: 589M Active, 209M Inact, 146M Wired, 512K Cache, 111M Buf, 46M Free
> Swap: 1200M Total, 1200M Free
> 
>   PID USERNAME    THR PRI NICE   SIZE    RES STATE    TIME   WCPU
> COMMAND
>   927 root          1  97    0 52084K 28172K select   1:02  2.98% Xorg
>  1128 vova          1  96    0 21172K 13676K select   0:10  0.98%
> metacity
>  1252 vova          1  96    0 38024K 29884K select   1:56  0.63%
> skype_bin
>  1386 vova         11 126    0   516M   470M RUN      3:42  0.00%
> evolution-2.6
>  1134 vova          3  20    0 21540K 13496K kserel   1:38  0.00%
> gkrellm
>  1113 vova          1  96    0 19592K  9604K select   0:17  0.00%
> at-spi-registryd
>  1327 vova          1  96    0 34464K 24668K select   0:10  0.00% sim
>  1323 vova          1  96    0 24552K 17212K select   0:09  0.00%
> cpufreq-applet
> ...
> 
> 
> No any idea who eats these 95.5% of CPU time. 
> Same picture on vmstat's pigs screen.
> 
> I know that this process is actually evolution, and if I kill it system
> load drops, but why it is not shown by top (and other).

As I understand it if the application is working in a thread then the cpu
statistics are not available. So a threaded program can often be using
100% cpu and show up as barely any in top. Then you have to try and
guess which one of the processes it likely is.

I get this a lot on my desktop and it is very annoying. As far as I know it
cannot be easily fixed.

> Any hints about it ?
> 

Thanks,
Dominic