GNOME Bugzilla – Bug 573115
[udpsink] set TTL and LOOP properties for multicast groups
Last modified: 2009-08-31 10:19:40 UTC
In commit e206f74bcea198673e6e64f04218d7c72a56efcd from 2008-05-21 19:35:50 (gst/udp/: Joining a multicast group and setting the loop/ttl properties are totally unrelated tasks are must be separ...), the functionality of gst_udp_join_group() was split in two functions. However, the body of the second (new) function - gst_udp_set_loop_ttl() - was set in an '#if 0 .... #endif' block. The result is that multicast streams have a default (multicast) TTL of 1 and only have significance in your own subnet. Result: your streams will not pass a single router. I have two patches which I will attach shortly: 1- Make sure the TTL and loop properties are set when multicasting 2- Match the default multicast TTL to the (more sane?) default value of 1
Created attachment 129486 [details] [review] Set the TTL and loop properties for multicast groups
Created attachment 129487 [details] [review] Set default multicast TTL value to 1
Commited the patch in Comment #1. Why again do you want to set the default TTL to 1? Your previous comment seems to imply that a default TTL of 1 only makes sense in your own subnet and won't pass a single router, which seems a rather limiting default to me. commit 474d9d7a9b9b32bfd66a558be468e7a5b588f992 Author: Jan Smout <jan dot smout at gmail dot com> Date: Thu Feb 26 13:19:31 2009 +0100 udp: fix gst_udp_set_loop_ttl() again Fix the gst_udp_set_loop_ttl() function that was commented out in a previous commit. See #573115.
(In reply to comment #3) Oops, I kind of forgot to motivate the second patch ... I'm following the reasoning as explained in RFC1112. Quoting the RFC (second paragraph of section 6.1 "Extensions to the IP Service Interface" on page 4): """ First, the service interface should provide a way for the upper-layer protocol to specify the IP time-to-live of an outgoing multicast datagram, if such a capability does not already exist. If the upper-layer protocol chooses not to specify a time-to-live, it should default to 1 for all multicast IP datagrams, so that an explicit choice is required to multicast beyond a single network. """ To my opinion, the udpsink implementation should also follow this recommendation. That said, another potential patch comes to mind: rename the udpsink ttl property to something like mc-ttl. This emphasizes that the property only affects multicast updsinks. One can go as far as to change the name throughout the sources for the same reason. This impacts gstmultiudpsink.c and gstudpnetutils.c. Just let me know if you're interested in this patch. Otherwise, I won't spend the effort.
Wim, what's your opinion on this?
*** This bug has been marked as a duplicate of bug 588245 ***