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

[Ramón García]

In terms of performance they are very similar. If UDP handles the
fragmentation itself, the big logical packet is lost if any of the
physical packets is lost. The same happens with RTP, because a packet
is an undivisible unit of decoding, and thus if a part is lost, then
one cannot decode anything. It could make sense to do some sort of
"bitpeeling", so that if the second part of a Vorbis packet is lost,
one can use the first to play an approximation of the intended sound.

With regard to fragmenting to overcome the limit of 64 Kb, I am
opposed to it if there is no significant use. In what case could
Vorbis packets be so large?

Ramon