Thanks for the answer. I hadn't thought about that. I wasn't using tunnels just to be sure PMTUD would not hit me (ouch !).
I am not sure Cisco IOS can load-balance on embeded IP packet as you say. Cheers, 2014-12-17 17:08 GMT+01:00 Saku Ytti <[email protected]>: > > On (2014-12-17 10:25 +0100), Xavier Nicollet wrote: > > Hey, > > > I know I could use > > show ip cef [VIP/32] detail > > show ip cef [VIP/32] internal > > or > > show mls cef exact-route [IPSRC] [VIP] > > > > However, I guess it would be easier to have exact internal hashing > > algorithm. > > Or is there another way to monitor each real server with such > configuration > > You probably want to have unicast address as well as anycast address and > NMS > the unicast address. > I don't think the hash algorithm is publically documented, as vendor > probably > does not want customers to rely on it not changing. > > I like this configuration, but there is one catch to it, it tends to make > PMTUD issues more pronounced, as there are no guarantees that the ICMP > message > generated by transit router will reach correct server, so it might cause > blackholing. > There are two cures for this, use smaller MTU on servers, which is > statistically unlikely to be too large for relevant portion of hosts. > Second, > prettier solution is to ask vendor to do ECMP hash for the embedded IP > packet > in ICMP message, instead of the top headers. > -- > ++ytti > _______________________________________________ > cisco-nsp mailing list [email protected] > https://puck.nether.net/mailman/listinfo/cisco-nsp > archive at http://puck.nether.net/pipermail/cisco-nsp/ > -- Xavier Nicollet _______________________________________________ cisco-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
