[dns-operations] [Ext] DNS Flag Day 2020 will become effective on 2020-10-01

[Tony Finch]

Paul Vixie <paul at redbarn.org> wrote:
>
> happily, that's not known about PLPMTUD (RFC 8899 & 8900). so right now
> there's new hope, yet undashed.

I'm trying to understand how PLPMTUD can help the DNS.

The DNS doesn't have a paketization layer as such - any time the DNS needs
to get a big message over a pipe that's too narrow, we get TC and fallback
to TCP which does the packetization for us.

My very superficial understanding is that PLPMTUD is based on
application-level probe / timeout / retry instead of ICMP errors. DNS
resolver implementations have machinery to probe the largest working EDNS
buffer size, so we already have something in the same ballpark as PLPMTUD.

For me the questions are:

  * what's the hard ceiling on a resolver's EDNS buffer size?

  * what's a resolver's starting probe buffer size?

  * how complicated is the resolver's probe algorithm?

  * what's the ceiling on an auth server's UDP response size?

Tony.
-- 
f.anthony.n.finch  <dot at dotat.at>  http://dotat.at/
Dover, Wight: Northeast 5 to 7. Moderate. Fair. Good.