Dear colleagues, After looking at the material presented in RFC 2374 and the discussions in the working group at the 38th RIPE meeting I am left with a few questions. My understanding of IPv6 is that it is supposed to solve some of the problems that we have with IPv4. These include: - Running out of IPv4 address space - Difficulties of Route aggregation - Renumbering And a new problem addressed at the last RIPE meeting - Running out of AS Numbers (prediction mid 2005) (It will become obvious later why I mention this here) With the current proposal, I do not see that any of these problems will be adequately (or if at all) addressed. The most obvious problem is that the 13 bits of sTLA space that is being given to providers today is TOO small. One way of increasing this would be to implement CIDR in IPv6, and not to include EUI-64 interface addresses in the IP header. Is arp really that bad that we have to throw it away? There is no medium that I know of that can support 2^64 addresses in the one segment, and I see no advantage of shipping this information across the Internet. Mutlihoming and Renumbering are still very much up in the air. Aggregation (including the GSE (8+8) draft) all seem to base their design on the Internet Registries (RIPE, etc ) or 'Tier 1' providers having the most significant bits (leftmost) in an IP address. The problem I see with this approach is that - It is not usual for an ISP to 'Peer' with a routing registry, and therefore not really getting a great gain from route aggregation - How do you determine which Tier a provider belongs to, and should he change size, moving or down a Tier is almost impossible due to renumbering difficulties. A suggestion that I would like to throw into the ring would be to include the AS Number into the start of the IP Header, and maybe a bits after that, to help control entry/exit points into boarding networks (a type of tag). | 3| 21 | 8 | 24 | 8 | 64 bits | +--+-----+---+--------+--------+--------------------------------+ |FP| AS |TAG| NLA | SLA | Interface ID | | | NUM | | ID | ID | | +--+-----+---+--------+--------+--------------------------------+ or | 3| 21 | 8 | 32 | 64 bits | +--+-----+---+-----------------+--------------------------------+ |FP| AS |TAG| NLA | Customers address | | | NUM | | ID | Range (CIDR) | +--+-----+---+-----------------+--------------------------------+ depending whether one wanted to get rid of arp. This way the border routers would only require the path information on how to get from one AS to the other, without requiring detailed information on the prefixes. Dual homing would be solved by issuing those customers with their own AS number, as is the practice today. It is obvious however that the number of bits in an AS Number would have to be increased, but this is the case anyway. To ease renumbering it might make sense to issue the numbers out of a smaller range (96 bit for example). The customers edge routers would then add the current ISPs AS Number to the packet before forwading them out. This way, the only change required when chaing ISPs is on the boarder routers. The field lengths/ names above are just an example, and would have to be closely examined. I would be interested in hearing any opinions, Regards Andrew -- Andrew Miehs Cybernet AG email: amiehs@cybernet-ag.net phone: +49 89 993 15-0 fax : +49 89 993 15-199