[xiph-rtp] Ident field and configuration - stream mapping

[Luca Barbato]

Since the Ident field has to be reconsidered a summary of what I 
remember now:

Implicit(Hash) map case:
- the crc32 may have collision problems, yet to be determined the how 
often and in which case, could be relatively easy to test feeding the 
encoder the same sample with different bitrates and see how the crc 
differ. I have no problem in believing that could happen.

Direct map case:
- one byte is enough for everyone. Is it true? is it false? Is 16bit 
enough? Is 32bit really big enough? Just take the worst case (and use IMHO):

a stream radio wants to just use one single continuous rtp stream.

with 256 codebook possible the stream would last about 1280m ( 5min per 
song, plausible) that means less than a day

take 65536 ,that means about 227 days

take 2^24 and you have enough years of music.

with 2^32 you have something more.

Let me just say that is an abusive behavior (better have different 
sessions) and as usual are random numbers.

Direct mapping will require to put a start value for the off band case 
or another way to map them correctly. I'm thinking either of additional 
fields or (ab)use the uri in a way similar to 
proto://host.x/path/stream_id/codebook_number

That said I hope that someone dislikes or likes my idea so we can come 
up with a possible solution.

My favorite so far:

24bit/16bit field with direct mapping


Off band : the url definition 
proto://host.x/path/stream_id/codebook_number make the association.

In band : is immediate since you can tag the Configuration Packet with 
the incremental map number, you notice corruption if the number doesn't 
match.

Inline: is easy too since you have a direct mapping by the order in 
which they are placed (the first is 0, the second 1 and so on).


lu

-- 

Luca Barbato

Gentoo/linux Developer		Gentoo/PPC Operational Leader
http://dev.gentoo.org/~lu_zero