David Botham wrote:
>
>
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> [clip...]
> > > The root servers do not maintain glue for .net name servers.
> > > That glue is in the gtld-servers.net servers.
> >
> > Yes, I know that, but my question was why not, and what about the
>
> I can only speculate as to the reason why the root name servers no
> longer host the Generic Top Level Domains. I am assuming it is due
> to load.
Absolutely, and .edu will likely move to a separate set of servers
soon. Again, the question is why do they not hold the glue for
nameservers when they are in a another TLD? I'm not worried about
those nameservers not being authoritative for that TLD.
> Your name server did not need to go back to the root name
> servers to find out where the .net name servers are unless this is
> the first query they have made (or the 6 day ttl on the ns records
> has expired).
Right, but if they are forced to go to the gTLD servers they will
generate two queries, one for each of the other nameservers, which
would have been unnecessary if the glue records had been there and
been handed back in the additional section of the first query.
These glue records are not really needed, but carry some benefit.
The benefit increases if the nameservers did not also happen to be
authoritative for a domain delegated by .net, because then other
intervening nameservers in the chain would have to be queried, not
just the gTLD servers. For example, the ns is ns1.b.c.net, but it
is not authoritative for c.net, then a query has to be generated to
the nameservers for c.net for ns1.b.c.net, etc. All for a query
in another TLD!
So, does this have to happen because, for example, BIND needs to
get those A RRs when it seeds its RTT data? Or does it not happen
until subsequent queries for that domain, and the RTT algorithm
force one of the other nameservers to surface?
> > effects of not doing so? And, is a querying nameserver steered
> > towards the nameserver for which it immediately gets the glue? Or
> > does it blindly go back to the root nameservers, and then the gTLD
> > servers to get it for the other two records, at least a couple of
> > extra queries? If the root nameservers are pounded for 10-15k
> > queries/sec with peaks around 30k/sec, getting only a partial
> > picture of authority from them seems silly, *even* though it is
>
> I think if you take a close look at the resolution process, you will
> see that the root servers are benefiting from the new arrangement. I
> think they answer a lot fewer questions by passing responsibility for
> .net off to the gtld name servers.
>
> Dave...
>
> > not strictly required. If it is not going to happen, it seems as
> > though I ought to be careful to make sure all of my authoritative
> > nameservers are in the same TLD.
Consider the last sentence above, and how things would be if .net
and .com were handled by separate sets of servers, *and* many ISPs
in the .net space were offering primary and/or secondary service
for their .com customers. About as efficient as a flyswatter with
a hinge in the middle!
-Robert
|
