|
|
|
|
@ -34,17 +34,30 @@ about performance
|
|
|
|
|
|
|
|
|
|
snprintf'ing and vsnprintf'ing all debug messages does consume
|
|
|
|
|
quite some CPU power, especially on call setup and tear down as
|
|
|
|
|
there are a lot of debug messages print out.
|
|
|
|
|
there are a lot of debug messages print out. for high cps with G711,
|
|
|
|
|
this can hinder good performance. some numbers:
|
|
|
|
|
|
|
|
|
|
for G711 announcement (from file) over loopback interface, the
|
|
|
|
|
following was measured over 2000 calls using sipp and valgrind:
|
|
|
|
|
|
|
|
|
|
call length CPU consumed by logging (log_fac_print, incl)
|
|
|
|
|
5s 34%
|
|
|
|
|
1s 58%
|
|
|
|
|
|
|
|
|
|
for 50 participants G711 conference, 30s/ call, logging
|
|
|
|
|
used 16%.
|
|
|
|
|
for GSM announcement cpu usage of encoding relativates the one of logging :
|
|
|
|
|
5s 9%
|
|
|
|
|
1s 14%
|
|
|
|
|
|
|
|
|
|
for iLBC announcement even more:
|
|
|
|
|
5s 1%
|
|
|
|
|
1s 4%
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
for participants conference, 30s/ call, logging
|
|
|
|
|
used
|
|
|
|
|
participants codec logging CPU
|
|
|
|
|
50 G711 16%
|
|
|
|
|
10 GSM 0.8%
|
|
|
|
|
5 iLBC 0.4%
|
|
|
|
|
|
|
|
|
|
valgrind command:
|
|
|
|
|
valgrind --tool=callgrind --instr-atstart=no ./sems -f sems.conf -D 1 -E
|
|
|
|
|
|