On 5/5/2015 1:34 PM, Dave Dolson wrote: > For IPv6, if the packet doesn't fit in the tunnel, the ICMP6 "too big" > message is sent, and the sender fragments it, sends it again, and notes > the PMTU for future packets. This is how IPv6 fragmentation is supported.
If the ICMP is received (and not blocked). If the ICMP has enough information for the tunnel source to know what happened. But this also then means that the tunnel egress will need to reassemble. > Some of your arguments are that implementations are broken so ICMP doesn't > work. I don't buy that. When the internet doesn't work, it gets fixed or > the customers go away. See RFC 4821. This issue has been known for a long time, and that's why the IETF developed an alternative that doesn't depend on ICMP receipt. > Your one argument I think may be valid is the IPv6 1280 requirement. > Let's say N levels of tunneling are required to reduce 1500 bytes to 1280 > bytes. > The person who wants to do the (N+1)th overlay either cannot do it, or they > resort > to fragmentation of the outer layer. > > Nonetheless, the approach of reducing Path-MTU is a valid 3rd choice, > provided it is > not reduced below 1280. The better solution would be RFC 4821-style probing by the tunnel ingress to the tunnel egress. Joe _______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
