option 212 for 6RD
achatz at forthnetgroup.gr
Sat Jan 19 16:22:23 CET 2013
Tassos Chatzithomaoglou wrote on 19/1/2013 16:27:
> Mikael Abrahamsson wrote on 19/1/2013 12:45:
>> Also, the references to the RFCs, that talks about MTU advertised in RA. I see no reason
>> why a host system can't have a per-route MTU and treat each RA received and the
>> connected routes created from it, have a different MTU.
>> I'm going to test this someday, I don't see a reason for it not to work.
> I don't know if technically MTU is prohibited to be used for per prefix assigment, but i
> don't see a problem having something (called "MTU", or whatever else) that pre-defines at
> the source the maximum packet size allowed to be used when ipv6 packets are sent towards a
> specific destination.
> In my mind the best scenario is that if the CPE knows the next link's "MTU" where the
> majority of prefixes are (WAN side), why not inform the hosts (LAN side) about it and
> prevent them from discovering it through PMTUD or any other "intrusive" way. I understand
> the possibility for the "MTU" to be further reduced in other hops, but we're comparing
> something with 100% probability (everything passing through the WAN must be lower than x)
> vs something with <100% probability (in further hops).
> In an old email of mine i had also proposed the idea of supplying an "MTU" per prefix
> through mif's dhcpv6 route option. Since some operating systems already store the PMTUD
> result in a per destination cache, why not pre-fill a part of that table?
> Having fought for years with IPv4 PMTUD issues, i don't see any real possibility to fix
> IPv6 ones, unless something of a MUST is introduced into the IPv6/ICMPv6 standard itself
> that enables this to work by default. And maybe it's too late for that too.
Just to add that Cisco's MRU in MPLS is actually MTU per FEC or prefix, because it informs
the LSR how big a packet can be for a particular FEC (i.e destination prefix) that can
still be forwarded out without being fragmented (the MRU changes per prefix depending on
the label operation performed on the prefix).
More information about the ipv6-ops