On Fri, Jun 15, 2012 at 12:09 PM, Timo Teras <timo.teras at iki.fi> wrote: > On Fri, 15 Jun 2012 11:46:50 +0700 Benny Prijono <bennylp at teluu.com> > wrote: > > > IOCP ioqueue has been deprecated and we haven't used it for long time. > > Please use select ioqueue. > > Why? IOCP is superior in performance by design. > > It's due to this: https://trac.pjsip.org/repos/ticket/1015 As I also noted, select() backend has also problems that make it bad > for me. > > See: > > http://lists.pjsip.org/pipermail/pjsip_lists.pjsip.org/2012-June/014742.html > > Right, thanks. I think the whole ioqueue thing is due for a make over. :) BTW. I would expect the memory corruption here happening exactly based > on what I observed: at least in transport_send_rtp() it is not checked > if the pending request is already in use - it'll just get reused > blindly, and will result in bad things if it was pending. > > Increasing MAX_PENDING might help a bit - but the proper fix is to have > the code check if pending queue is full and then make it start dropping > packets properly. > > And as noted, MAX_PENDING should be large enough to accommodate almost > the number of packets the sound card can send in one burst (about worth > half of the latency or so). > > That's true, ioqueue users (transport, in this case) is responsible for supplying free ioqueue_op_key to prevent the problem. Thanks Benny > -Timo > -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.pjsip.org/pipermail/pjsip_lists.pjsip.org/attachments/20120615/de8d32fe/attachment.html>
