[xiph-rtp] Fragmenting in RTP versus letting UDP fragment

[Tor-Einar Jarnbjo]

Onsdag, 3 november 2004, skrev Ramón García <ramon.garcia.f at gmail.com>:

>Even if retransmission has an important use case for streaming media,
>there isn't the necessary number of clients and servers supporting it
>to make it viable.

Splitting the Vorbis packets at application level makes at least 
then sence, if an application decides to make use of FEC to make 
the transport more reliable. Even if a fragment of a Vorbis packet 
is missing, it could be possible to reconstruct the page using the 
FEC data and the probability that this will succeed is higher, if 
just a part of the packet is missing.

As I stated earlier, some sort of packet splitting must be defined 
for Vorbis anyway, as a Vorbis packet may be larger than 64kB and 
the current Vorbis/RTP draft simply states that packets SHOULD be 
kept below the network path MTU size. This is not a strict restriction,
an implementation may choose to not care about the MTU size, so 
why have a long discussion about this subject?

Tor