[e2e] TCP ex Machina

Detlef Bosau mail at detlef-bosau.de
Fri Sep 20 15:32:03 PDT 2013

Am 20.09.2013 19:21, schrieb Daniel Havey:
> I think that the conversation is getting ramped up now ;^)
> |
> My thoughts on the paper are mixed.  I believe that centralized server calculated CC is the wrong direction for the future.  
Me too.

The more I think about it, the more I'm convinced that one of the most
fundamental design errors is to put CC on the end points.
This has become a religion and now, it is more likely that women become
priest in the roman catholic church than that this basic flaw will ever
be fixed.

> IMHO there are simply to many clients in network conditions that are too diverse, and too rapidly changing.  One size doesn't fit all.  I think that the client should provide the congestion control algorithm.

Have a look for the TEAR approach.

Actually, it doesn't make a difference whether you place CC at the
sender side or the receiver side - congestion occurs along the path and
hence must be fixed there - and not at the end points.
> So the paper takes a nice step in this direction.  

No. It runs into the same old vicious cycles which lead to nowhere in
the last 20 years.

> The client provides the network conditions to the server,

Daniel, the only one to provide network conditions to anyone is the
network - more precisely the node, where congestion occurs.
And I really wonder, why we did not think that way the last 20 years.

End points may report end point conditions. Network conditions must be
provided by the network. Not by endpoints, be it an educated guess, some
sophisticated conjecture or simply throwing dice.

More information about the end2end-interest mailing list