On 2012/04/20 08:54, Janne Johansson wrote: > Overall, this sounds like the "missing arp" issue in some regards.
That was specifically affecting 50+ carp interfaces with the same physical parent interface (or iirc bridged vlans). > As for the recent threads about carps misbehaving, I had at least one > pair of carped servers that worked better when I defined static > carppeers and stopped using multicast. > Yes, its a workaround, but it made the carps reliable for me. Yes sometimes there can be a problem if switches are partially or incorrectly configured for multicast however one would still expect to see outgoing packets on the interface. Might be interesting to see all of Matt's netstat -s, not just the carp section. Actually I would set advbase 0 to increase the number of carp packets generated per second, record 'netstat -s' and then another one after a couple of seconds, and send a diff of the two. > If you can Matt, try to set the carppeer option so it unicasts carp > status packets between the hosts over the vlans, and see if it helps. The parent iface not being in promisc mode is likely to at least break reception of packets destined for the carp MAC address (unless bnx(4) can program the MAC filter independent of promisc, I don't have bnx on any boxes that I can play with).