PROBLEM: Preemptible kernel makes mpg123 skip a lot under 2.6.0-testing7and very high load average under low usage.

From: Dru
Date: Thu Nov 06 2003 - 18:04:17 EST


Con Kolivas wrote:

Hi.

I quote from your output:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
22953 andru 15 0 10100 5316 9464 S 3.7 0.6 2:02.39 mpg123
1067 root 5 -10 595m 58m 539m S 3.3 6.6 391:41.29 XFree86
1176 andru 15 0 47488 26m 13m S 1.0 3.0 11:52.32 gnome-terminal
25063 root 17 0 2004 1096 1792 R 0.7 0.1 0:00.03 top

The kernel is now tuned to give much more priority to reniced tasks and it is not recommended to run your X server nice -10. This is the cause of your problem as X is starving your audio application. Some distributions do this by default to get around the limitations of the old cpu scheduler not being able to make X smooth enough at nice 0. This hack/workaround is no longer recommended for 2.6 kernels. You will find nice performance of X at nice 0 now and audio will not skip when the nice value of X is the same as your audio application.

Con



Hello all. i've still been testing the schedular for test7. Sorry havn't tried out testing9 yet, me and
the computer have been having some emotional differences recently. Plus rebooting is a long processes
breaks my ps/2 mouse support (USB mouse fine) and sometimes my modelM keyboard.

If i run 10 low cpu intensive processess at once (that reguarlly require wakeup many times a second,
but chew less than 1% of the cpu) they will starve other processors a lot. I get mouse jerkness
when the system is running many processes with 80% cpu usuage or higher. Though the major
problem is with mplayer, and mpg123. Renicing seems to have little effect on the schedular.
I could renice mpg123 to -19, start up 10 processors that run in the background that operate at nice
of 10 and I will still have sound skip (not as much as not nicing processes at all).
mpg123 doesn't do buffering ahead as xmms does.

I will say this, the stablity of 2.6.0-testing is very good in regard to uptime. I'm finding it better than
2.4 in some respect.





top - 11:50:58 up 4 days, 17:31, 44 users, load average: 1.27, 1.43, 1.52
Tasks: 197 total, 1 running, 196 sleeping, 0 stopped, 0 zombie
Cpu(s): 23.8% us, 3.3% sy, 1.3% ni, 22.8% id, 45.0% wa, 0.0% hi, 3.6% si
Mem: 904792k total, 899952k used, 4840k free, 1080k buffers
Swap: 1914656k total, 253764k used, 1660892k free, 492360k cached


PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
10327 root 15 0 651m 62m 538m S 18.6 7.1 408:38.62 XFree86
31958 andru 0 -19 9412 3732 9048 S 3.3 0.4 0:58.05 mpg123
10404 andru 15 0 46952 23m 15m S 2.3 2.7 19:40.27 gnome-terminal
322 root 15 0 0 0 0 S 0.7 0.0 5:10.32 rpciod
16268 andru 25 10 18492 7084 4736 S 0.7 0.8 6:11.64 btdownloadcurse
16584 andru 25 10 33172 5928 4736 S 0.7 0.7 3:02.38 btdownloadcurse
32274 root 17 0 2212 1096 1868 R 0.7 0.1 0:00.03 top
10384 andru 15 0 11312 5668 9844 S 0.3 0.6 1:37.27 metacity
28693 andru 26 10 18320 7468 4736 S 0.3 0.8 8:16.47 btdownloadcurse
16145 andru 15 0 18628 7060 4736 D 0.3 0.8 3:11.18 btdownloadcurse
16334 andru 30 15 18452 6984 4736 S 0.3 0.8 4:29.09 btdownloadcurse
16420 andru 15 0 26056 7344 4736 S 0.3 0.8 3:41.03 btdownloadcurse
32093 root 15 0 3184 1604 2936 S 0.3 0.2 0:04.48 http



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