On Jan 14, 2015, at 6:54 AM, Tony Finch <[email protected]> wrote:
> 
> Paul Hoffman <[email protected]> wrote:
>> 
>> From an operational standpoint "an auth server can put more answers in
>> than the querier might expect" is a heck of a lot easier to deploy than
>> "teach the stubs to ask better questions".
> 
> True. And if authoritative servers include more answers, recursive servers
> also need to be taught to pass the additional data to their clients.

Although the draft doesn't say it explicitly, I think the optimization is only 
"fill the recursive's cache faster". This reduces latency on the current flow 
of requests from the stub, but does not reduce the flow itself.

> An alternative and much easier way to get Warren's performance improvement
> is for browsers to remember which domain names were needed by a website.
> Then for later visits the browser can do the subordinate lookups in
> advance, in the hope that the list of domains hasn't changed much.

I'm pretty sure some browsers have been doing this for years. My interest in 
the draft is for non-Web requests that have multiple relevant answers, such as 
SRV. Even there, however, I suspect the increase in speed to the stub will be 
miniscule.

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

Reply via email to