In einer eMail vom 10.08.2010 08:32:10 Westeuropäische Sommerzeit schreibt [email protected]:
On Aug 9, 2010, at 8:07 AM, Joel M. Halpern wrote: > There is an interesting question of how to handle congestion management when we move the multi-pathing down to the network layer. > On the one hand, architecturally it seems appropriate for network layer to be managing the path selection. > > On the other hand, the congestion behavior of the various paths is likely to be quite different, and congestion response is definitely NOT the network layer's responsibility. > > The correct balance is not obvious to me. (This is similar to, but not the same as, the path selection policy issues that the MIF working group is trying to understand. I'd recommend that we also consider locator selection as another degree of freedom. 'Right. I agree with Joel that flow control belongs to the transport layer, and that path selection is part of the network layer. Right. Is locator selection a transport function? It would seem that having it be a unilaterally network layer function is over constraining the architecture. Yet, transport cannot perform intelligent locator selection without network layer input. Consider locator selection to be the host's next hop selection. Also consider the possibility that a congested node propagates in upstream direction (wrt to the congestion causing incoming flows) intelligent notification information ( degree of congestion: one goal might be to avoid congestion in the first place; distance i.e.(transitively incremented) number of hops down to the issuing congested node,....). The more upstream node may more easily select an alternative next hop which ensures bypassing the congested node - than the immediate neighbor node e.g. Only eventually the upstream going congestion notification will reach the sending host as to induce a change of locator selection. Yes, transport cannot perform intelligent locator selection without network layer input. It starts with intelligent data provisioning of an intelligent network, which is forwarded upstream - up to the sending host eventually. Heiner
_______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
