In einer eMail vom 10.08.2010 23:54:23 Westeuropäische Sommerzeit schreibt [email protected]:
For example, if multiple locator pairs all result in paths that coincide on a particular congested link, then the benefits of multi-pathing could be extremely limited. This is exactly one of the points I have been trying to make. Detouring the congested node/cluster should start at some optimal closest node - at best there where the alternate path is not really a detour. A detour path that starts at the sending host may easily melt with the original path prior reaching the congested node. It may even cause additional problems in case the path between ingress and "melt"-node is known to be "less performing" (lesser SLA). And: MPTCP's intelligent locator selection can never be a concerted action. Note, a congestion is not caused by one single IP-flow. Nor is it caused by a completely broken node (unless indirectly), which means that some flows should continue to pass the congested node while others should detour it. I can only repeat referring to that concept which I tried to launch 6 years ago but was rated poor (assumably by some RRG subscribing expert) : I am still sure that it is the right way to go - no matter what MPTCP or CONEX are performing. Heiner
_______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
