21 Dec
2006
21 Dec
'06
1:43 p.m.
At 12:00 21/12/2006, Bogdan-Andrei Iancu wrote:
Jiri Kuthan wrote:
Well -- I think the assessment language needs to be made more accurate. The bug which
you reported is a bug and I thank you for the report.
The problem I'm referring to though, is a major system inefficiency. The timer system
in question is low-resolution, inaccurate, invariable and slow -- simply opposite of
real-time. I personally believe that the results of the performance comparison are
attributable to it (note that this is a guess though) but again, that's only one of
more defficiences.
-jiri
--
Jiri Kuthan http://iptel.org/~jiri/
if you happen to have a PCAP file with the
incident, let me please know.
-jiri
p.s. even if you didn't tweak timers, the results may be suboptimal because
the software version you are using is having rather indeterministic timer subsystem.
For example, the recent measurements (http://www.iptel.org/ser/doc/performance)
show quite scattered server responsiveness under high load. (Note though that
the measurement results were achieved in a best-effort manner based on the tester
knowledge and understanding of openser and that the result are not officially confirmed by
the OpenSER project.) Whether it is indeed the cause is not certain though
-- this looks really like a stealth bug.
it is not a bug - the timer's behaviour is quite difficult to predict on real
traffic platforms and experience showed that it has nothing to do with the patterns
observed during lab stress tests. Personally, I'm very keen when it comes to finding
bugs and as a matter of fact I did submitted a full bug report for the timers in the
latest SER stable version, couple of weeks ago. I also checked and this bug was not
present in non of the OpenSER versions.