To:
Havard Eidnes <he@runit.no>
Cc:
randy@psg.com, seamus@bit-net.com, users@ipv6.org, dnsop@cafax.se, ngtrans@sunroof.eng.sun.com
From:
"Perry E. Metzger" <perry@wasabisystems.com>
Date:
19 Jan 2001 19:27:28 -0500
In-Reply-To:
Havard Eidnes's message of "Sat, 20 Jan 2001 01:20:13 +0100"
Sender:
owner-dnsop@cafax.se
Subject:
Re: IPv6 dns
Havard Eidnes <he@runit.no> writes: > > We further assume that no root server machines return AAAA or > > A6 records for machines serving "." unless it gets queried over > > v6 transport, in order to minimize the "can't fit in a > > datagram" issue. > > You'll have to excuse my inexperience with the intimate details > of BINDv9, but is this code already part of BINDv9? No, but from the looks of things, it should straightforward. > I don't suppose use of EDNS0 would serve as a suitable alternative > distinguishing criterion for whether to add v6 glue records to > replies? The issue is not the vintage of the resolver -- it is a UDP datagram length problem. We assume a much larger v6 datagram can get through without the risk of fragmentation. Such a v6 datagram is reasonably large at 1280 bytes. The v4 datagrams have much less space for v6 glue records, since we can only assume 512 bytes of payload before fragmentation. (The truth is it would probably be okay if we assumed most of the world had a much larger MTU even for v4. That's a whole different issue, though.) > > > o its ip address will be cached in other servers > > > > ...which is okay since they're real servers with perfectly fine data... > > ...and no new IPv4 address would be added for the new "separate" > v6-serving hosts, I suppose. Correct. -- Perry E. Metzger perry@wasabisystems.com -- Quality NetBSD CDs, Support & Service. http://www.wasabisystems.com/