On Friday 13 February 2009, toly wrote:
We are running the test:
hammer->sbc(asterisk 1.4)->openser1.3.3->mediaserver(heavily modified asterisk)
It is production environment.
Hi Toly,
thanks for the feedback.
call rate is steady 4 cps.
Proxy is 2 dual cores 2.5Ghz with 4G of ram
Initially openser was compiled with private memory size of 4M. After few days privated memory was heavily fragmented and proxy was running out of memory. Then I've compiled with 8M of ram. It prolonged more but with the same result, which made me very nervous and PKG_MALLOC was out of the question.
Hm, this is strange. We using 1.3 PKG_MALLOC in a production environment with heavy traffic, and don't any fragmentation instabilities. You're not the first that reports this, thought.
Without PKG_MALLOC:
There were few runs for few days, the last one since last friday still running. No instabilities on proxy, no problems at all. Performance wise I have not notice any difference. It may be subjective, I'd say cpu may be higher (%0.1 sometimes) but again it's subjective and I'm running ngrep logging via syslog_ng and the proxy logging done via syslog_ng, configured (syslog_ng) to send the output to the syslog_ng server and pumping all huge output via syslog_ng may contribute to the CPU.
Good to know that its aparently don't cause any problems.
BTW, in main.c when opens syslog, there is the thing which may be qualified as bug. I you use standard syslog then there is no problem. With syslog_ng, there is the problem: it creates 2 log files and I've fixed main.c, by the code it writes in syslog, and then when daemonized opens syslog.
It creates two log files? We've also use syslog-ng in some instances, never noticed a problem. What exactly you need to change?
In turn, I'd like to ask the question about ngrep. I'm using ngrep-1-45. When running on proxy I see duplicate message. When running not on proxy - no dups. My hunch is that it's recording all stuff from on_reply_route and loose_route?
I don't think its related to the OpenSER, perhaps this is something in the network configuration of this machine? This could e.g. happens when you use bonded interfaces, and traces on the "any" interface.
Cheers,
Henning