[Home]

Summary:DAHTOOL-00050: [patch] dahdi_test accurately reports nonsense
Reporter:Matthias Urlichs (smurfix)Labels:
Date Opened:2011-01-04 06:22:07.000-0600Date Closed:2011-09-11 00:15:59
Priority:MinorRegression?No
Status:Closed/CompleteComponents:Utilities
Versions:2.4.0 Frequency of
Occurrence
Related
Issues:
Environment:Attachments:( 0) 0001-dahdi_test-Enforce-range-from-0.0-100.0-for-accuracy.patch
Description:dahdi_test reports -200% or 400% accuracy.

That's clearly stupid.

Also, the percentage printed in verbose mode clearly doesn't match the one printed in terse mode.

SVN trunk as of last week.

****** ADDITIONAL INFORMATION ******

root@lab# dahdi_test
Opened pseudo dahdi interface, measuring accuracy...
-200.014% -200.008% -200.012% -200.011% -200.012% -200.012% -200.013% -200.013%
-200.013% -200.014% -200.008% -200.012% -200.013% -200.013% -200.011% -200.013%
-200.012% -200.012% ^C
--- Results after 18 passes ---
Best: 0.000 -- Worst: -200.014 -- Average: -200.012032, Difference: 400.012032
root@lab:/var/lib/dkms/dahdi/2.3+svn-smurf2/build# dahdi_test -v
Opened pseudo dahdi interface, measuring accuracy...

8192 samples in 32769.188 system clock sample intervals (400.014%)
8192 samples in 32768.566 system clock sample intervals (400.007%)^C
--- Results after 2 passes ---
Best: 0.000 -- Worst: -200.014 -- Average: -200.010704, Difference: 400.010705


root@lab:/var/lib/dkms/dahdi/2.3+svn-smurf2/build# lsmod
Module                  Size  Used by
dahdi_transcode         5542  0
dahdi_dummy             2437  0
dahdi                 186362  2 dahdi_transcode,dahdi_dummy
Comments:By: Shaun Ruffell (sruffell) 2011-01-04 10:57:21.000-0600

Something is wrong on your system:

With a default build today:

[root@foxtrot dahdi-linux]# /etc/init.d/dahdi stop
Unloading DAHDI hardware modules: done
[root@foxtrot dahdi-linux]# modprobe dahdi
[root@foxtrot dahdi-linux]# cat /sys/module/dahdi/version
SVN-trunk-r9614
[root@foxtrot dahdi-linux]# dahdi_test& sleep 6; killall dahdi_test
[1] 9841
Opened pseudo dahdi interface, measuring accuracy...
100.000% 99.996% 99.999% 99.999% 99.999% [1]+  Terminated              dahdi_test
[root@foxtrot dahdi-linux]# dahdi_test -v & sleep 6; killall dahdi_test
[1] 9870
Opened pseudo dahdi interface, measuring accuracy...

8192 samples in 8191.967 system clock sample intervals (100.000%)
8192 samples in 8191.648 system clock sample intervals (99.996%)
8192 samples in 8191.888 system clock sample intervals (99.999%)
8192 samples in 8191.913 system clock sample intervals (99.999%)
8192 samples in 8191.888 system clock sample intervals (99.999%)[1]+  Terminated              dahdi_test -v
[root@foxtrot dahdi-linux]#


If I increase the chunksize from 8 to 40 the accuracy of any specific period goes down, but with the -v option you can see that over time it's balanced around where it should be.

[root@foxtrot dahdi-linux]# /etc/init.d/dahdi stop
Unloading DAHDI hardware modules: done
[root@foxtrot dahdi-linux]# modprobe dahdi
[root@foxtrot dahdi-linux]# cat /sys/module/dahdi/version
SVN-trunk-r9614M
[root@foxtrot dahdi-linux]# dahdi_test& sleep 6; killall dahdi_test
[1] 10804
Opened pseudo dahdi interface, measuring accuracy...
99.903% 99.606% 99.904% 99.903% 99.903% [1]+  Terminated              dahdi_test
[root@foxtrot dahdi-linux]# dahdi_test -v& sleep 6; killall dahdi_test
[1] 10833
Opened pseudo dahdi interface, measuring accuracy...

8192 samples in 8199.977 system clock sample intervals (100.097%)
8192 samples in 8159.704 system clock sample intervals (99.606%)
8192 samples in 8199.896 system clock sample intervals (100.096%)
8192 samples in 8199.904 system clock sample intervals (100.096%)
8192 samples in 8199.896 system clock sample intervals (100.096%)[1]+  Terminated              dahdi_test -v
[root@foxtrot dahdi-linux]#


Are you running dynamic spans or something?  What is the output of dahdi_scan?
By: Shaun Ruffell (sruffell) 2011-01-04 11:03:31.000-0600

(but yes...I'll agree that the non-verbose mode should be reporting 400% in your case, and not a negative percentage)
By: Matthias Urlichs (smurfix) 2011-01-04 14:23:31.000-0600

# dahdi_test -v& sleep 6; killall dahdi_test

Hmm, isn't that what "dahdi_test -c5" is supposed to do? ;-)

Anyway:

root@lab:~# dahdi_scan
[1]
active=yes
alarms=UNCONFIGURED
description=DAHDI_DUMMY/1 (source: HRtimer) 1
name=DAHDI_DUMMY/1
manufacturer=
devicetype=DAHDI Dummy Timing
location=
basechan=1
totchans=0
irq=0
root@lab:~#

So it seems that a broken patch in my build script mistakenly enabled dahdi_dummy, and that dahdi_dummy is currently not only out-of-date but broken. My bad.

root@lab:~# rmmod dahdi_dummy
root@lab:~# dahdi_test -v -c5
Opened pseudo dahdi interface, measuring accuracy...

8192 samples in 8191.384 system clock sample intervals (99.992%)
8192 samples in 8190.648 system clock sample intervals (99.983%)
8192 samples in 8191.256 system clock sample intervals (99.991%)
8192 samples in 8191.128 system clock sample intervals (99.989%)
8192 samples in 8191.136 system clock sample intervals (99.989%)
--- Results after 5 passes ---
Best: 99.992 -- Worst: 99.983 -- Average: 99.989207, Difference: 99.989208

OK, so now please fix the "Difference:" part of the result. :-P
By: Shaun Ruffell (sruffell) 2011-01-04 16:19:54.000-0600

I just dropped the "Difference" field in the attached past...and bounded the range from 0 - 100% in the attached 0001-dahdi_test-Enforce-range-from-0.0-100.0-for-accuracy.patch

I also added a "cummulative" accuracy when running in verbose mode to smooth out jitter in each pass (larger chunksizes == more jitter).

Thoughts?


By: Shaun Ruffell (sruffell) 2011-04-18 10:35:50

@smurfix: Are you going to test this?