RE: Regarding IPv6 prefix filtering
Your list is already very strict. As we have discussed yesterday, it might be desireable (until there is a better solution for multihoming) to explicitely permit prefixes up to a /48. This could be done "for your own region only", or "for all regions",
I am personally opposed to allow /48 PA prefixes in the routing table. This would become a de-facto the end of aggregation and lead us quickly to an un-manageable routing table again that would take huge efforts to clean. Although I have sometimes disagreed with Pim, the filters he posted are the responsible thing to do. We have reached this situation of people requesting 6bone pTLAs or subTLAs for the sole purpose of being multihomed because too many people have put their head in the sand for too long. Please don't encourage this anymore. /48s in the routing table are not a solution. There are two different IPv6 multihoming solutions currently discussed on the ipv6mh list that deal with geographic addresses (with a finer graining than a region) that could be aggregated. Michel.
Hi, On Thu, May 02, 2002 at 08:34:12AM -0700, Michel Py wrote:
I am personally opposed to allow /48 PA prefixes in the routing table. This would become a de-facto the end of aggregation and lead us quickly to an un-manageable routing table again that would take huge efforts to clean.
Currently there is no other working and implemented way to get v6 multihoming. Multiple /48s obviously do not work (think failover of established TCP sessions and routing policy decisions), regional addressing isn't here (and I'm unsure how that is supposed to work concerning non-regionality of upstream networks), so what else do we have? * "PI" (get me my own allocation) -> *this* is creating things that will *never* go away * announce a /48 out of an aggregate (and accept that people that are "far away" may opt to filter all more-specifics from other regions, or even from *all* regions) -> if we figure out that this doesn't work, these announcements can be filtered, and we have neither created legacy "PI" networks nor any special policies that we'll have a hard time getting rid of. This is why I think that *today* (and for the next year or so), for companies that we want to be multihomed (like the RIPE NCC), /48s are a workable solution. There is no other *today*, and to get things like "build a DNS chain with v6 transport" going, you need it *now*. Gert Doering -- NetMaster -- Total number of prefixes smaller than registry allocations: 47584 (44543) SpaceNet AG Mail: netmaster@Space.Net Joseph-Dollinger-Bogen 14 Tel : +49-89-32356-0 80807 Muenchen Fax : +49-89-32356-299
participants (2)
-
Gert Doering -
Michel Py