[dccp] Is dropping flows enough to avoid congestion collapse?

[Eddie Kohler <kohler@icir.org>]

There have been suggestions that dropping flows is enough to avoid
congestion collapse, and that RTP-over-UDP already does this.

Well, sort of.  Currently, *users* may drop flows when they see high levels
of loss.  On the other hand they may not.  Certainly we can't count on it,
and we can't count on users watching for any particular loss rate.  1%?
30%? 40%?  Depends on the codec and the user.

Carrying through the IAB's suggestion would mean endpoint operating
systems, or perhaps routers or middleboxes, terminating UDP flows, or
treating UDP as less than best effort, when there is congestion.

That would seem a deeply nonsatisfying result, and probably far worse for
user experience than DCCP.  (I'm also not sure DCCP will be that bad --
there are several misunderstandings floating around, as well as some real
issues.)

Of course, UDP-penalization might not ever happen, but some "people working
on the stability of the network" want it to.  (I'm thinking of Sally
Floyd.)

Eddie

_______________________________________________
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