lkml.org 
[lkml]   [2004]   [Nov]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [patch] Real-Time Preemption, -RT-2.6.10-rc1-mm3-V0.7.23
Date
>I may do a variant on this anyway. I think its important to see if
>the symptom (> 100 usec CPU delay) is really:
>- lots of short delays
>OR
>- relatively few long delays
>and I have an idea of how to code that up and add to latencytrace.

A follow up on this message. My first test completed with the following
results. The new code indicates:

X - Min delay was 0. Max delay was 3. Ave delay was 0.015295.
top - Min delay was 0. Max delay was 23. Ave delay was 0.025659.
netO - Min delay was 0. Max delay was 31. Ave delay was 1.169024.
netI - Min delay was 1. Max delay was 35. Ave delay was 1.182864.
diskW - Min delay was 0. Max delay was 18. Ave delay was 1.166944.
diskC - Min delay was 0. Max delay was 18. Ave delay was 1.080036.
diskR - Min delay was 0. Max delay was 7. Ave delay was 0.803804.

A "delay" was counted if 1000 iterations of the CPU inner loop took
longer than 10 usec. For comparison, I moved this code to my 2.4 system
and got the following results:

X - Min delay was 0. Max delay was 17. Ave delay was 1.277730.
top - Min delay was 0. Max delay was 12. Ave delay was 1.452692.
netO - Min delay was 0. Max delay was 12. Ave delay was 1.633742.
netI - Min delay was 0. Max delay was 12. Ave delay was 1.626565.
diskW - Min delay was 0. Max delay was 14. Ave delay was 1.566188.
diskC - Min delay was 0. Max delay was 12. Ave delay was 1.701542.
diskR - Min delay was 0. Max delay was 12. Ave delay was 1.650909.

Looks pretty comparable at this level. The 2.4 results appear to be more
consistent.

Grr. The new code does have an impact on the application measurements
under 2.4. It appears the TSC accesses are being delayed while the disk
is active. The within 100 usec rate was only 77% (was 90%) but the peak
is still pretty close (2.60 vs. 2.38 msec).

I am also not sure this is the "right" measurement either. I probably
need to count the delays or divide the overall loop time by the number
of delays to see if that is a more meaningful value.

--Mark

-
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/

\
 
 \ /
  Last update: 2005-03-22 14:07    [W:0.038 / U:3.488 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site