> Date: Thu, 24 Jun 2010 16:11:17 +0700 > From: bennylp@xxxxxxxxx > To: pjsip at lists.pjsip.org > Subject: Re: High load after upgrade to release 1.5.5 (and 1.6) > > On Tue, Jun 15, 2010 at 7:11 PM, Arie Velthoen > <arie_velthoen at hotmail.com> wrote: > > Hi, > > > > For your info: > > > > It appears that PJSUA release 1.5.5 (and 1.6) on WinXP/2003 results in a > > much higher cpu-load than before (release 1.5 and back), when having many > > RTP-channels/calls configured. > > > > This appears to be caused by ticket #1015 Disable ioqueue Completion Port > > backend > > > > I understand that release builds of pjproject 1.5 and older: use in PJLIB > > ioqueue_winnt.c > > > > whereas in release 1.5.5 and up this is replaced by ioqueue_select.c which > > appears to be much less efficient. > > > > I have 2048 RTP-channels configured. > > > > My config_site.h contains: > > > > # define PJSUA_MAX_CALLS 2048 > > > > # define PJSUA_MAX_CONF_PORTS 4098 > > > > # define PJSUA_MAX_PLAYERS 2048 > > > > # define PJ_IOQUEUE_MAX_HANDLES 4096 > > > > Yes we had to disable IOCP ioqueue due to unsolved bug, as the ticket says. > > Do you really need to configure for that many calls? I mean to be > frank I wouldn't expect PJSUA-LIB to be able to handle 2048 calls, > because it was not designed that way (as the FAQ says, > http://trac.pjsip.org/repos/wiki/FAQ#pjsua-lib-perf). And unlike the > IOCP, the extra/unused calls may bring with them some polling > overhead, so it's best to reserve the least amount that you're going > to use. > > Cheers > Benny > > _______________________________________________ > Visit our blog: http://blog.pjsip.org > > pjsip mailing list > pjsip at lists.pjsip.org > http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org Thank you for your response. My intention was to use about 500-1000 calls, but it appeared that there was no performance penalty in raising PJSUA_MAX_CALLS to 2048 (before release 1.5.5). On a "Xeon 5130 at 2.00GHz 2G Win2003R2" system I have reached some 1200 calls with a wav-player each, using SIPp with 40 calls/sec and 30 sec hold-time. Knowing the above remark from the faq, this was a surprise to me also. I noticed that the original IOCP bug ticket #985 is closed as being "fixed". Knowing there is such a difference in performance I am still interested in a real solution to this bug. Unfortunately I have no idea how to attack this bug myself. So any suggestion is welcome. Regards, Arie _________________________________________________________________ Een netbook met Windows 7? Hier vind je alles dat je moet weten. http://www.windows.nl/netbook -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.pjsip.org/pipermail/pjsip_lists.pjsip.org/attachments/20100624/302395d7/attachment.html>
