On Sat, Aug 7, 2010 at 7:26 PM, RJ Atkinson <[email protected]> wrote: > I am not sure what you mean above.
Exactly what you are describing below, i.e. the session is established between the identifiers and you have three simultaneously paths between identifiers > > The ILNP implementation can send TCP segments (or TCP ACKs, > or other TCP information) using different Locator pairs. > > As an example, if one considers TCP with an originator using > Locators (A, B, C) and a responder using Locators (W, X, Y), > then the originator can choose which Source Locators to use > and also which Destination Locators on a packet-by-packet > basis. > > Again, purely as an example, the originator could send > packets using these Locator values in this simple sequence: > (A, W) > (B, X) > (C, Y) > or any other sequence that it wishes to. Similarly, the > responder can use any valid combination of Locators that it > wishes to use. > > In any case, the TCP implementations at either end are unaware > that multiple Locators are being used (i.e. because the > transport-layer pseudo-header checksum only includes Identifiers, > never Locators). In turn, this is why not special multi-path > TCP (or UDP or SCTP or other) transport-layer modifications > are required. (Caveat: Of course, the ILNP stack upgrade is > needed in the first place.) > > The same concepts apply to UDP and/or SCTP. > > I will look over the ILNP I-Ds and try to clarify this a bit; > apparently they aren't clear enough right now in this area. There are some challenges ahead though a) if e.g. A-W path gets congested, how to detect the congestion and switch load over to the other paths, see http://datatracker.ietf.org/doc/draft-ietf-mptcp-congestion/ b) A, B, C and W,X,Y are describing where the border routers are installed between the edge and core networks. If there are middleboxes in some of the paths then some mechanism is needed (e.g. multi-homed hosts) to ensure you have symmetric flows through the border routers (it is likely the middlebox is installed close to the border router). c) should this multipathing feature belong to the network layer, i.e. creating e.g. a generic congestion mechanism for all transport protocols or is multipathing something that the transport protocol layer should take care of? It might be valuable to have some discussion around c) before you spend your time on updating your I-Ds Also, many thanks for the clarification! -- patte _______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
