From: Ben Schwartz <[email protected]>
Sent: Monday, January 25, 2021 5:04 PM
To: Hollenbeck, Scott <[email protected]>
Cc: [email protected]; [email protected]
Subject: [EXTERNAL] Re: Re: Re: [DNSOP] SVCB without A/AAAA records at the 
service name







On Mon, Jan 25, 2021 at 8:06 AM Hollenbeck, Scott 
<[email protected]<mailto:[email protected]>>
 wrote:

      [SAH] We need to think about the impact on the servers that have to 
respond to those queries, too. Sending unnecessary queries to the root and TLD 
servers puts a load on those servers that can have an impact on every 
client/resolver that needs to be able to reach them.



   This is important, but I don't think it's applicable here.  The various 
options under consideration all impose the same amount of load on root and TLD 
servers.

   [SAH] So if some number if queries X and some number of queries Y are 
processed in parallel, the value of X + Y will be the same as if those queries 
are processed serially?



   Yes, if the SVCB record uses the hostname as the TargetName, as suggested in 
section 10.2, then there are no additional queries; they are merely issued in 
parallel rather than sequentially.



   Regardless of whether the speculative queries are wasted, your question was 
whether they would impose more load on TLD or root servers.  If the domain's NS 
record is cached at the resolver, then they certainly will not.  (All queries 
will go to the authoritative nameserver.)  If QNAME minimization is applied, 
they also should not.  (They will all trigger the same qname-minimized query to 
the root and TLD, and a reasonably intelligent resolver shouldn't emit 
duplicate queries when one is already in flight.)



   If the NS record is not cached, and the resolver does not implement qname 
minimization, then perhaps it is possible that the additional queries could 
leak to the TLD, or to the root if the TLD NS record is not cached.  The 
behavior would be similar to speculative AAAA queries today.  (I'm not aware of 
these being a cause for alarm among root or TLD operators.)



   Note that this conversation only concerns hypothetical future non-HTTP 
protocols that rely exclusively on SVCB.  We are a very long way from any such 
protocol (1) existing and (2) having enough usage to concern the root or TLD 
operators.

   [SAH] Thanks for the clarification, but I have to disagree that it’s not a 
concern for operators. We would be wise to pay attention to such things long 
before use causes measurable impact on those servers.



   Scott

_______________________________________________
DNSOP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dnsop

Reply via email to