On Tue, Feb 3, 2015 at 12:19 PM, John W. Linville <linvi...@tuxdriver.com> wrote:
> Thanks for doing that research for me! :-) You don't see any way to > work with Dave M. around his objections? > It seems to be a philosophical objection, which is likely intractable. > Violating an "upstream first" policy seems like you are making an "end > run" in hopes of eventually coming back to Dave M. with an argument > like "but people have been using this already". People are already using it, so some sufficiency of deployment probably isn't going to tip the scales.. :-) Fundamentally this comes down to a disagreement over whether users should be allowed to choose to use a protocol which performs poorly or not at all in the face of certain topologies. Personally I believe that users ought to be given that choice to make for themselves (and this is really already true at varying levels for all of the supported tunneling protocols, due to device support, firewalls, middleboxes, etc.), but I don't think you're going to find community consensus on this. As an aside, the assertion that routers absolutely depend on tools like RED is patently ridiculous. It is possible that naive *networks* depend on these tools, but the router couldn't care less. TCP back-off and shaping tools that depend on knowledge of the state machine are effectively "cooperative" bandwidth management tools that hope - but cannot guarantee - that they know how the sender will react and you should *never* rely on them contributing significantly to traffic engineering on your network. -- Nick
_______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
