On Tue, May 4, 2010 at 11:13 PM, Bernd Eckenfels <bernd-10@xxxxxxxxxxxxx> wrote: > In article <u2t73dbe2851005041257l9e2cdf50sab187fd49b4688c7@xxxxxxxxxxxxxx> you wrote: >> 1) GO = 1->2->4 >> RETURN = 4->1 >> 2) GO = 1->3->4 >> RETURN = 4->1 > > if the two interfaces at 1 have different ip addresses, it will result in an > auto bind of the socket to one of the addresses. If you use a fixed local > address it might be due to the fact that its not the FIB which is working > here, but the tcp socket endpoint doing the routing decision. Node 1 has the same IP address at each interface. I add that I see the same strange behaviour if I do send the packets with UDP. (I tried this with "nc -u") Anyway, this sounds new to me. Is the socket endpoint storing the routing decision for future transmissions? Maybe also a UDP endpoint? So, if I get it right, when a socket is connected to an address the lookup is performed and the next-hop is memorized; then when that socket is used the next-hop will always be the same. How am I supposed to do effectively multipath then? -- To unsubscribe from this list: send the line "unsubscribe linux-net" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
