[MUD-Dev] Re: Error tolerant UDP data streams

[Ola Fosheim Grøstad <olag@ifi.uio.no>]

J C Lawrence wrote:

> --<cut>--
> The extreme compression achieved by the LPC-10 algorithm allows the
> option of ``robust transmission,'' in which multiple copies of sound
> packets are sent, each containing a sequence number which allows the
> receiver to discard duplicate or out-of-sequence packets. Robust
> transmission often allows intelligible conversation over heavily
> loaded network links which would otherwise induce random pauses and
> gaps in received sound.
> --<cut>--

I haven't seen this, but straight duplication is not a very good idea for
sound. (Is LPC linear predictive coding?)

>
> it thru in good order.  Bad connection/lag == more duplicates and
> more network congestion.  Good connectivity == minimal duplication.

I've been thinking about this a while back and have pretty much reach the
same conclusion. To phrase it differently: if I am the one filling up the
bottlenecks then brute force duplication is a very bad and expensive idea. I
think a better approach is to classify events and make vital events very
small so they can be duplicated by appending them to other messages.
Reducing the number of vital events sent is of course a good design goal...

A better approach, for "non-boolean" data in which low fidelity is valid
term, would be to interleave the information in various packages. Thus, if
you get 50% of the packages then you will get the "full" signal at half
fidelity.

As for the firewall problem mentioned in another message. I believe one
should be able to bypass that to some extent by using several TCP-IP
connections and use OOB cleverly?  Anyway, one can always send UDP-style
messages down a TCP-IP pipe, so designing the protocol for UDP is not a
total waste.
--
Ola