On 11/01/2010 10:29 EDT, Joel M. Halpern wrote: > I actually think that the question of where one should put end-point > identification and where one should put path agility is somewhere more > complicated than just assuming that we should put a session layer on top > of transport. > While that approach works, it has drawbacks. > One example of the drawback is that the way MPTCP is architected, since > it is using vanilla TCP, it can not switch which path it is using for a > piece of data it has already sent unless that path is declared > completely dead. retransmission have to use the same path as the > original (or else you stall that connection.
I don't want defend MPTCP, but actually that bit of data can be retransmitted on another path. Because of some middlebox behavior, it will be retransmitted on the broken path as well until the problem is resolved, but the endpoint can know to trash it. As for recent messages in general, I get the feeling that we probably need to revisit the concepts of "session" and "session layer", and maybe what each layer provides in the way of end-to-end state. Maybe in the halls. _______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
