Hi,

I certainly agree that we should not block ourselves from better transport in 
the future, therefore it is important that 
the following SHOUDs are SHOULDs and definitely no MUSTs: 

DNS servers SHOULD NOT relay on path MTU discovery or PLPMTUD (RFC4821/RFC8899) 
but 
SHOULD use IPV6_USE_MIN_MTU=1 from RFC 3542 to avoid the need to do path MTU 
discovery.

With the current believed state of ICMP / MTU blackholing, I fear it is the 
best we can get working right now for UDP,
hopefully we can reise this in 10 years and increase the recommendation.

With regards to TCP & QUIC… I am unsure whether PLPMTUD has any chance with 
short-lived flows to discover anything.
Perhaps we should as @Gorry for his expert opinion on that one…

AVE!
   Philipp


> On 13. Aug 2025, at 12:01, Joe Abley <[email protected]> 
> wrote:
> 
> Hi Paul,
> 
> It seems to me that there are only two approaches here:
> 
> (a) standardise a way for DNS messages over UDP/IPv6 to be retained after 
> sending in some predictable fashion so that they can be fragmented at source 
> if necessary, or
> 
> (b) specify that when sending a DNS message over stateless transport where 
> path MTU discovery is expected (over UDP over IPv6) the sender of the 
> response MUST NOT send a message so large that fragmentation is possible.
> 
> Personally I think that (a) is simply reinventing the overhead of using 
> stateful transports. Focusing on a wholesale move to those transports and 
> giving up on large UDP messages is better than trying to bolt state as an 
> afterthought onto UDP.
> 
> 
> Joe
> 
>> On 13 Aug 2025, at 11:14, Paul Vixie <[email protected]> 
>> wrote:
>> 
>> Are we using the recent fragmentation avoidance rfc so soon? Mark, the 
>> locally witnessed network MTU does not completely predict the full path MTU. 
>> Plpmtud is and should remain in scope. Let's not close off a way to reach 
>> reasonable datagram sizes in the future. Every 10x in speed should someday 
>> show up as 3.333x in packet rate and 3.333x in packet size. We are using 
>> tinygrams in comparison. What we do today will affect the possibilities 
>> later. Vixie
>> Paul Vixie
>> Aug 5, 2025 03:30:00 Mark Andrews <[email protected]>:
>> 
>> draft-ietf-dnsop-3901bis-03 states: 
>> 
>>       If the requesting resolver is unable to process fragments, or if 
>>       fragments are filtered on-path, resolution will fail over UDP. 
>>       These issues are more prevalent for IPv6, as it no longer allows 
>>       on-path hosts to fragment packets. Therefore, working Path MTU 
>>       Discovery (PMTUD) is essential for IPv6 DNS-over-UDP packets to be 
>>       fragmented to a size that allows them to traverse all segments on 
>>       a path. 
>> 
>> This is not factually correct.  There is NO requirement to perform PMTUD 
>> at all in a DNS server over IPv6.  For UDP you just fragment at network 
>> MTU in the sending node at network MTU.  This can be achieved by using 
>> IPV6_USE_MIN_MTU=1 socket option from RFC 3542 or using an interface that 
>> is configured with a MTU that matches the network MTU.  For TCP you use 
>> the socket option TCP_MAXSEG to set the MSS to the network MTU. 
>> 
>> Both of these options have been used for years in nameservers and avoid 
>> response losses caused by attempting PMTUD itself. 
>> 
>> -- 
>> Mark Andrews, ISC 
>> 1 Seymour St., Dundas Valley, NSW 2117, Australia 
>> PHONE: +61 2 9871 4742              INTERNET: [email protected] 
>> 
>> _______________________________________________ 
>> DNSOP mailing list -- [email protected] 
>> To unsubscribe send an email to [email protected]
>> _______________________________________________
>> DNSOP mailing list -- [email protected]
>> To unsubscribe send an email to [email protected]
> 
> _______________________________________________
> DNSOP mailing list -- [email protected]
> To unsubscribe send an email to [email protected]

AVE!
   Philipp S. Tiesel

-- 
Philipp S. Tiesel 
https://philipp.tiesel.net/

_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to