On Aug 19, 2010, at 3:50 PM, Ralph Droms wrote: > Being a little pedantic here...my understanding is that a host never knows a > subnet length, per se. What the host knows is a list of on-link prefixes, > which it matches against outbound traffic. A minimal implementation might > not keep a list of on-link prefixes and send everything to its default > router; redirects would do the rest...
If it goes to the default router, it will still get there (but maybe ttl--; as a side-effect). > > - Ralph, who is sorry if he missed something earlier in the thread and is now > pedanting up the wrong tree The reason I'm being pedantic about this is we have had network outages as a result of redirects being on by default as a hidden configuration option. I've similarly seen the same problem in ipv4 w/ redirects + proxy-arp taking out the device or driving the cpu load high enough that it impacts other protocls, etc.. Overall, while a "nice-to-have", it does damage network stability and SHOULD NOT be enabled by default. this is why we put it in our interface templates to disable this capability. Here's how many ipv6 interfaces we have enabled today with no adverse customer or infrastructure impact after this change. cfgtools=> select count(*) from ipv6; count ------- 1600 - Jared -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------