On Wed, 23 Oct 2002, Matthew Toseland wrote:

> On Wed, Oct 23, 2002 at 12:09:06AM +1000, fish wrote:
> > 
> > Another day, another different test of the streaming stuff:
> > 
> > http://www.dspaudio.com/~jaymz/freenet-audio-streamer-test9.tar.gz  
> > http://www.dspaudio.com/~jaymz/freenet-encoder.zip  
> > http://www.dspaudio.com/~jaymz/streamserver.zip
> > 
> > The encoder is much more robust now.  yay.  I was able to stream for an
> > hour using 10 second slices, and i didn't run into any problems with 60
> > second slices
> Interesting. Between two different permanent nodes or just locally?

one permanent, one transient

> Given freenet's latency problems, I suspect the minimum that will work
> for the whole network will be some sort of FEC scheme - make each minute
> (or larger time unit) a redundant splitfile (with really small chunks - 
> fetch several of them at once, keep fetching until you have finished the
> minute. Standard FEC tools such as fishtools can probably handle it).
> Otherwise, you will have to buffer a long way ahead to work around worst
> case download times, which can be pretty horrendous on freenet.

FEC is next on my list, actully.  The major issue is this, tho: the time
to insert/retrieve a piece of data is *not* linear compared to it's
size.  That is to say, it only takes 1.2x the time to insert/retrieve a
256k chunk that it does to insert/retrieve a 16k chunk.  You cna alieviat
this somewhat by throwing threads at the problem, however, you can't throw
more than 25-30 at freenet, or else your node will go into overload
mode (even though many of those threads aren't actully doing any
processing ;)).

I'll probably make the defaults to be 60 second normal chunks, and 10
second metachunks or something like that.  But it won't happen until the
weekend, most likely, since I've got a whole bunch of work to catch up on.

        - fish


_______________________________________________
devl mailing list
devl at freenetproject.org
http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to