Re: 2.4.17 absurd number of context switches

Davide Libenzi (davidel@xmailserver.org)
Fri, 28 Dec 2001 10:06:35 -0800 (PST)


On Fri, 28 Dec 2001, Jeffrey W. Baker wrote:

>
>
> On Fri, 28 Dec 2001, Alan Cox wrote:
>
> > > Check out those figures for context switches! 30,000 switches per second
> > > with only three runnable processes and practically no block I/O seems
> > > quite high to me. You can also see that the system is spending half its
> > ..
> > > Is this a scheduler worst-case, something to be expected, or something I
> > > can work around?
> >
> > The scheduler is _good_ at the three process case. Run some straces it looks
> > more like postgres is doing wacky yield based locks.
>
> All I see in strace is semop forever
>
> [pid 10076] 0.000054 semop(1179648, 0xbfffe6e8, 1 <unfinished ...>
> [pid 10077] 0.000224 <... semop resumed> ) = 0
> [pid 10077] 0.000077 semop(1179648, 0xbfffe1e8, 1) = 0
> [pid 10077] 0.000057 semop(1179648, 0xbfffe0f8, 1 <unfinished ...>
> [pid 10076] 0.000128 <... semop resumed> ) = 0
> [pid 10076] 0.000035 semop(1179648, 0xbfffe6a8, 1) = 0
> [pid 10076] 0.000127 semop(1179648, 0xbfffe758, 1 <unfinished ...>
> [pid 10077] 0.000085 <... semop resumed> ) = 0
> [pid 10077] 0.000075 semop(1179648, 0xbfffe0f8, 1) = 0
> [pid 10077] 0.000155 semop(1179648, 0xbfffdfb8, 1 <unfinished ...>
> [pid 10076] 0.000401 <... semop resumed> ) = 0
> [pid 10076] 0.000034 semop(1179648, 0xbfffe758, 1) = 0
> [pid 10076] 0.000046 semop(1179648, 0xbfffe758, 1 <unfinished ...>
> [pid 10077] 0.000113 <... semop resumed> ) = 0
> [pid 10077] 0.000040 semop(1179648, 0xbfffdf78, 1) = 0
> [pid 10077] 0.000051 semop(1179648, 0xbfffdfc8, 1 <unfinished ...>
> [pid 10076] 0.000317 <... semop resumed> ) = 0
> [pid 10076] 0.000055 semop(1179648, 0xbfffe718, 1) = 0
> [pid 10076] 0.000083 semop(1179648, 0xbfffe8d8, 1 <unfinished ...>
> [pid 10077] 0.000217 <... semop resumed> ) = 0
> [pid 10077] 0.000091 semop(1179648, 0xbfffdfc8, 1) = 0
> [pid 10077] 0.000057 semop(1179648, 0xbfffdfa8, 1 <unfinished ...>
> [pid 10076] 0.000191 <... semop resumed> ) = 0
> [pid 10076] 0.000037 semop(1179648, 0xbfffe898, 1) = 0
> [pid 10076] 0.000054 semop(1179648, 0xbfffe928, 1 <unfinished ...>
> [pid 10077] 0.000056 <... semop resumed> ) = 0
> [pid 10077] 0.000034 semop(1179648, 0xbfffdf68, 1) = 0

It's not the a sys_sched_yield() problem. probably and IPC_NOWAIT issue

- Davide

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/