[Bug 750456] rtpsession: Implement sending of reduced size RTCP packets

GStreamer (GNOME Bugzilla) bugzilla at gnome.org
Fri Jun 5 08:45:13 PDT 2015


https://bugzilla.gnome.org/show_bug.cgi?id=750456

--- Comment #2 from Sebastian Dröge (slomo) <slomo at coaxion.net> ---
Some IRC discussion around this:

<slomo_> wtay: https://bugzilla.gnome.org/show_bug.cgi?id=750456 do you have an
opinion? whether caps for negotiation, or a property? (or both, but no)
<wtay> slomo_, if it's in the SDP, caps
<wtay> slomo_, property not so much..
<slomo_> wtay: how would it look like in the caps? a=rtcp-rsize becomes
rtcp-rsize=true?
<wtay> slomo_, a-rtcp-rsize=1 or something, I think it starts with a-
<slomo_> wtay: ok, i'll check the code :) we also have trr-int, which should
override the rtcp-min-interval property btw...
<wtay> slomo_, but it's possible that it doesn't like attributes without =
<slomo_> wtay: there's also the problem that you would probably like to
override rtcp-rsize from the application, if you detect that something in the
middle is dropping the packets
<slomo_> so we probably also need a property additionally, maybe with an enum
that takes 3 values ;)
<wtay> slomo_, maybe
<ocrete> slomo_: shouldnt rtpsession try to detect if something is dropping the
packets ?
<slomo_> ocrete: that's not really trivial unfortunately. how do you detect
that a fir was successful? only the depayloader knows, or the decoder
<slomo_> but ideally yes
<ocrete> yes, it's indeed non-trivial, I hoped you had smart ideas ;)
<slomo_> that's why i was thinking of putting this responsibility to whatever
is around rtpbin, so that it can check if keyframes are arriving, etc :)
<wtay> it looks like a=rtcp-rsize will result in a caps entry a-rtcp-rsize and
an empty string
<slomo_> ocrete: we could add something to the depayloaders to have them send
an event upstream whenever they detect that some rtcp thing they did was
successful or not ;)
<ocrete> and rtx
<slomo_> that's something rtpjitterbuffer could detect, yes
<slomo_> ocrete: so basically we could have a backlog of event seqnums in
rtpsession for rtcp requests, and requesters should tell us with another event
when they were successful or timed out
<ocrete> btw, I think it shuld be a property... the rule I tried to use was, if
it's per payload, put it in the caps, if it's per session, as a property
<slomo_> makes sense

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.


More information about the gstreamer-bugs mailing list