[dccp] RE: Would DCCP kill Internet real-time media distribution?

["Rosen, Brian" <Brian.Rosen@marconi.com>]

> > Is the sense of the avt group really that slow start is completely
> > unacceptable for any important media flows?
> 
> Also, for what it's worth, I do understand and sympathize 
> with the problems
> that slow start can cause a truly-CBR application.  I don't think
> slow-start is all peaches and cream.
> 
> The problem is, Best Effort Public Internet == Slow Start.
Where does it say that?

Right now, we send UDP.  It actually works pretty well, but it
clearly doesn't work if you have a bunch of TCP sources gobbling
up the bandwidth.  DCCP is not going to change the effect of TCP
on the media, it's going to change the effect of the media on
TCP.  We don't need no DCCP.  "Fair" to media is all or nothing,
Philippe's ideas notwithstanding.  Nothing is fair if the alternative 
is garbage. 

If your goal is to prevent what is today mythical congestion
collapse, then we would be much better served by talking
about how signalling can get loss rates so that it can shut
down flows.  That's simple, and it won't screw anything up.

Brian

> 
> There are proposals to eliminate the need for slow-start 
> through explicit
> feedback from routers (XCP, QuickStart), but they're a ways 
> off.  Maybe
> media will be the app that pushes them through Cisco?  Significant TCP
> changes are probably even less likely.  (They're certainly less likely
> without any proposal on the board.)
I'll agree with this, although I'm also willing to work on solutions.

> 
> In the meantime, we really do have a situation where a media 
> application
> should either do slow start or something similar, or it 
> should stay off the
> public internet.  Brian, I'm coming to the conclusion that 
> your app should
> stay off the public internet -- or you may need to change the 
> app so that
> it will work best off the public internet, when it can get a 
> true bandwidth
> reservation.  Maybe that's OK for how your app is used in practice?
Sorry, but as you might imagine, I don't agree. See the note from
Henning.  We use 4M, while other systems might use 768K.  That's no
reason to have any difference in protocols.  Slow start is not
a practical thing for interactive use, sorry.

_______________________________________________
dccp IETF mailing list: dccp@ietf.org
list info:  https://www1.ietf.org/mailman/listinfo/dccp
wg charter: http://www.ietf.org/html.charters/dccp-charter.html