Hi Carlos,

On 8 Jul 2025, at 13:57, Carlos Horowicz <[email protected]> 
wrote:

> That said, a simple A record lookup typically involves multiple sessions to 
> different authoritative servers if you follow the chain from the root — not 
> one long session.

That depends a bit on where you are looking.

Between a stub and a resolver there are often lots of messages exchanged 
between a small number of hosts. You could imagine a long-held session between 
a stub and a resolver being used to exchange many messages.

If the resolution graph features forwarders between stub and resolver or 
between resolvers, you can also easily imagine many messages being exchanged 
between the same pair of hosts, in which case a long-held session might be 
useful.

When it comes to messages exchanged between resolvers and authoritative 
servers, some authoritative servers might be interesting enough that a resolver 
often has messages to send (e.g. a COM server, or an authoritative server that 
serves frequently-neededresponses with low TTLs). Long-held sessions might be 
plausible in those scenarios. Other authoritative servers might be reached more 
occasionally, in which case the idea of a long-held session seems less useful. 
The cost of setup/teardown for the latter might well be higher than some of the 
other scenarios, but if it happens only occasionally perhaps that's ok.

I once wrote up some ideas about long-held sessions and allowing endpoints to 
give hints to the other end about how long they might be kept open, which were 
ultimately published in RFC 7828. I remain unconvinced about how useful any of 
my ideas were (but surely the content added by those co-authors responsible for 
pushing the languishing draft over the finishing line is great).

> In contrast, sessions where large responses are expected (like NSEC or NSEC3 
> proofs) will almost certainly happen over TCP.

Be careful about what you predict. Real life is often more weird than you 
expect.

> Just sharing some field thoughts — thanks for your insight.


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

Reply via email to