[e2e] characterizing video conferencing traffic..

[Thierry Turletti]

> (Spencer_talking_out_his_hat)
> 
> It looks like this paper is putting one H.261 frame in each UDP
> packet, because I think I can derive the pie chart on packet
> size distribution from the repeating pattern of full-sized
> I-frames, separated by shorter P-frames and B-frames, modulo
> RTCP overhead.

Actually with H.261 a frame can contain both INTRA (I) and INTER (P) -encoded
macroblocs  (B frames is used in H.263/++). Such a mix is more efficient than 
Full INTRA encoding which generates very large frames and may introduce jitter 
in the decoding.

Whatever the codec used in this paper, packet size generated strongly depend 
on the details to encode in each image and on the movements in the scene to 
encode.

Thierry