Re: [patch] Real-Time Preemption, -RT-2.6.10-rc1-mm3-V0.7.23

From: Mark_H_Johnson
Date: Wed Nov 10 2004 - 09:25:31 EST


>> [4] Application level latencies are OK but not great.
>> X test - only 90% of CPU loops are within 100 usec of nominal value.
>> In previous RT kernels I got > 99% with 100 usec.
>
>this might be a side-effect of the chrt-ing of events/[0|1] and/or
>ksoftirqd (which we did to debug the 'freeze' problems) - are those
>still chrt-ed?
For reference:
# ps -eo pid,pri,rtprio,cmd | grep '\['
1 23 - init [5]
2 139 99 [migration/0]
3 34 - [ksoftirqd/0]
4 34 - [desched/0]
5 139 99 [migration/1]
6 34 - [ksoftirqd/1]
7 34 - [desched/1]
8 41 1 [events/0]
9 41 1 [events/1]
10 34 - [khelper]
15 32 - [kthread]
27 34 - [kblockd/0]
28 34 - [kblockd/1]
36 24 - [khubd]
103 23 - [kswapd0]
104 32 - [aio/0]
105 33 - [aio/1]
180 139 99 [IRQ 8]
195 14 - [kseriod]
201 139 99 [IRQ 12]
237 139 99 [IRQ 14]
239 139 99 [IRQ 15]
278 139 99 [IRQ 1]
310 24 - [kirqd]
313 139 99 [IRQ 4]
320 24 - [kjournald]
605 139 99 [IRQ 10]
1206 24 - [kjournald]
1207 24 - [kjournald]
1309 139 99 [IRQ 3]
1323 31 - [IRQ 7]
1494 139 99 [IRQ 6]
1748 139 99 [IRQ 11]
14131 23 - [pdflush]
14242 24 - [pdflush]
17337 21 - grep \[

>Please review and double-check all SCHED_FIFO tasks in
>the system and keep only those that are absolutely necessary for
>latencytest's operation [i.e. the soundcard IRQ and latencytest itself]
>- everything else should be SCHED_OTHER. Do latencies get any better if
>you do this?
I can, but that is not necessarily an "apples to apples" comparison.
When I compare with 2.4 preempt + low latency kernels, the X stress
test had > 99% of the samples within 100 usec of the nominal value.
Don't forget - on a 2.4 kernel, the IRQ's are all unthreaded. On
the 2.4 kernel, heavy disk I/O is where I get the worst behavior
and even then, I get > 90% of samples within 100 usec.

I still maintain that a 2.6 RT kernel has to do as well or better
than a 2.4 RT kernel (or else, why would I step up??).

--Mark

-
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/